Lines Matching +full:page +full:- +full:size
10 * Released under the terms of 3-clause BSD License
16 * struct page(s) to form a zspage.
18 * Usage of struct page fields:
19 * page->private: points to zspage
20 * page->freelist(index): links together all component pages of a zspage
21 * For the huge page, this is always 0, so we use this field
23 * page->units: first object offset in a subpage of zspage
25 * Usage of struct page flags:
26 * PG_private: identifies the first component page
27 * PG_owner_priv_1: identifies the huge component page
66 * span more than 1 page which avoids complex case of mapping 2 pages simply
72 * A single 'zspage' is composed of up to 2^N discontiguous 0-order (single)
101 #define _PFN_BITS (MAX_POSSIBLE_PHYSMEM_BITS - PAGE_SHIFT)
105 * encoding <page, obj_idx> and the encoded value has a room
116 * header keeps handle which is 4byte-aligned address so we
121 #define OBJ_INDEX_BITS (BITS_PER_LONG - _PFN_BITS - OBJ_TAG_BITS)
122 #define OBJ_INDEX_MASK ((_AC(1, UL) << OBJ_INDEX_BITS) - 1)
137 * On systems with 4K page size, this gives 255 size classes! There is a
138 * trader-off here:
139 * - Large number of size classes is potentially wasteful as free page are
141 * - Small number of size classes causes large internal fragmentation
142 * - Probably its better to use specific size classes (empirically
150 #define ZS_SIZE_CLASSES (DIV_ROUND_UP(ZS_MAX_ALLOC_SIZE - ZS_MIN_ALLOC_SIZE, \
184 * We assign a page to ZS_ALMOST_EMPTY fullness group when:
204 * Size of objects stored in this class. Must be multiple
207 int size; member
217 static void SetPageHugeObject(struct page *page) in SetPageHugeObject() argument
219 SetPageOwnerPriv1(page); in SetPageHugeObject()
222 static void ClearPageHugeObject(struct page *page) in ClearPageHugeObject() argument
224 ClearPageOwnerPriv1(page); in ClearPageHugeObject()
227 static int PageHugeObject(struct page *page) in PageHugeObject() argument
229 return PageOwnerPriv1(page); in PageHugeObject()
234 * For every zspage, zspage->freeobj gives head of this list.
242 * It's valid for non-allocated object
288 struct page *first_page;
325 pool->handle_cachep = kmem_cache_create("zs_handle", ZS_HANDLE_SIZE, in create_cache()
327 if (!pool->handle_cachep) in create_cache()
330 pool->zspage_cachep = kmem_cache_create("zspage", sizeof(struct zspage), in create_cache()
332 if (!pool->zspage_cachep) { in create_cache()
333 kmem_cache_destroy(pool->handle_cachep); in create_cache()
334 pool->handle_cachep = NULL; in create_cache()
343 kmem_cache_destroy(pool->handle_cachep); in destroy_cache()
344 kmem_cache_destroy(pool->zspage_cachep); in destroy_cache()
349 return (unsigned long)kmem_cache_alloc(pool->handle_cachep, in cache_alloc_handle()
355 kmem_cache_free(pool->handle_cachep, (void *)handle); in cache_free_handle()
360 return kmem_cache_alloc(pool->zspage_cachep, in cache_alloc_zspage()
366 kmem_cache_free(pool->zspage_cachep, zspage); in cache_free_zspage()
400 static int zs_zpool_malloc(void *pool, size_t size, gfp_t gfp, in zs_zpool_malloc() argument
403 *handle = zs_malloc(pool, size, gfp); in zs_zpool_malloc()
404 return *handle ? 0 : -1; in zs_zpool_malloc()
454 MODULE_ALIAS("zpool-zsmalloc");
457 /* per-cpu VM mapping areas for zspage accesses that cross page boundaries */
462 return zspage->isolated; in is_zspage_isolated()
465 static __maybe_unused int is_first_page(struct page *page) in is_first_page() argument
467 return PagePrivate(page); in is_first_page()
470 /* Protected by class->lock */
473 return zspage->inuse; in get_zspage_inuse()
479 zspage->inuse += val; in mod_zspage_inuse()
482 static inline struct page *get_first_page(struct zspage *zspage) in get_first_page()
484 struct page *first_page = zspage->first_page; in get_first_page()
490 static inline int get_first_obj_offset(struct page *page) in get_first_obj_offset() argument
492 return page->units; in get_first_obj_offset()
495 static inline void set_first_obj_offset(struct page *page, int offset) in set_first_obj_offset() argument
497 page->units = offset; in set_first_obj_offset()
502 return zspage->freeobj; in get_freeobj()
507 zspage->freeobj = obj; in set_freeobj()
514 BUG_ON(zspage->magic != ZSPAGE_MAGIC); in get_zspage_mapping()
516 *fullness = zspage->fullness; in get_zspage_mapping()
517 *class_idx = zspage->class; in get_zspage_mapping()
524 zspage->class = class_idx; in set_zspage_mapping()
525 zspage->fullness = fullness; in set_zspage_mapping()
529 * zsmalloc divides the pool into various size classes where each
532 * classes depending on its size. This function returns index of the
533 * size class which has chunk size big enough to hold the give size.
535 static int get_size_class_index(int size) in get_size_class_index() argument
539 if (likely(size > ZS_MIN_ALLOC_SIZE)) in get_size_class_index()
540 idx = DIV_ROUND_UP(size - ZS_MIN_ALLOC_SIZE, in get_size_class_index()
543 return min_t(int, ZS_SIZE_CLASSES - 1, idx); in get_size_class_index()
550 class->stats.objs[type] += cnt; in zs_stat_inc()
557 class->stats.objs[type] -= cnt; in zs_stat_dec()
564 return class->stats.objs[type]; in zs_stat_get()
589 struct zs_pool *pool = s->private; in zs_stats_size_show()
599 "class", "size", "almost_full", "almost_empty", in zs_stats_size_show()
604 class = pool->size_class[i]; in zs_stats_size_show()
606 if (class->index != i) in zs_stats_size_show()
609 spin_lock(&class->lock); in zs_stats_size_show()
615 spin_unlock(&class->lock); in zs_stats_size_show()
617 objs_per_zspage = class->objs_per_zspage; in zs_stats_size_show()
619 class->pages_per_zspage; in zs_stats_size_show()
623 i, class->size, class_almost_full, class_almost_empty, in zs_stats_size_show()
625 class->pages_per_zspage, freeable); in zs_stats_size_show()
652 pool->stat_dentry = debugfs_create_dir(name, zs_stat_root); in zs_pool_stat_create()
654 debugfs_create_file("classes", S_IFREG | 0444, pool->stat_dentry, pool, in zs_pool_stat_create()
660 debugfs_remove_recursive(pool->stat_dentry); in zs_pool_stat_destroy()
683 * For each size class, zspages are divided into different groups
687 * status of the given page.
696 objs_per_zspage = class->objs_per_zspage; in get_fullness_group()
711 * Each size class maintains various freelists and zspages are assigned
723 head = list_first_entry_or_null(&class->fullness_list[fullness], in insert_zspage()
727 * Put pages with higher ->inuse first. in insert_zspage()
731 list_add(&zspage->list, &head->list); in insert_zspage()
735 list_add(&zspage->list, &class->fullness_list[fullness]); in insert_zspage()
746 VM_BUG_ON(list_empty(&class->fullness_list[fullness])); in remove_zspage()
749 list_del_init(&zspage->list); in remove_zspage()
754 * Each size class maintains zspages in different fullness groups depending
756 * objects, the fullness status of the page can change, say, from ALMOST_FULL
758 * a status change has occurred for the given page and accordingly moves the
759 * page from the freelist of the old fullness group to that of the new
786 * to form a zspage for each size class. This is important
790 * usage = Zp - wastage
791 * where Zp = zspage size = k * PAGE_SIZE where k = 1, 2, ...
793 * For example, for size class of 3/8 * PAGE_SIZE, we should
800 /* zspage order which gives maximum used size per KB */ in get_pages_per_zspage()
809 usedpc = (zspage_size - waste) * 100 / zspage_size; in get_pages_per_zspage()
820 static struct zspage *get_zspage(struct page *page) in get_zspage() argument
822 struct zspage *zspage = (struct zspage *)page->private; in get_zspage()
824 BUG_ON(zspage->magic != ZSPAGE_MAGIC); in get_zspage()
828 static struct page *get_next_page(struct page *page) in get_next_page() argument
830 if (unlikely(PageHugeObject(page))) in get_next_page()
833 return page->freelist; in get_next_page()
837 * obj_to_location - get (<page>, <obj_idx>) from encoded object value
839 * @page: page object resides in zspage
842 static void obj_to_location(unsigned long obj, struct page **page, in obj_to_location() argument
846 *page = pfn_to_page(obj >> OBJ_INDEX_BITS); in obj_to_location()
851 * location_to_obj - get obj value encoded from (<page>, <obj_idx>)
852 * @page: page object resides in zspage
855 static unsigned long location_to_obj(struct page *page, unsigned int obj_idx) in location_to_obj() argument
859 obj = page_to_pfn(page) << OBJ_INDEX_BITS; in location_to_obj()
871 static unsigned long obj_to_head(struct page *page, void *obj) in obj_to_head() argument
873 if (unlikely(PageHugeObject(page))) { in obj_to_head()
874 VM_BUG_ON_PAGE(!is_first_page(page), page); in obj_to_head()
875 return page->index; in obj_to_head()
900 static void reset_page(struct page *page) in reset_page() argument
902 __ClearPageMovable(page); in reset_page()
903 ClearPagePrivate(page); in reset_page()
904 set_page_private(page, 0); in reset_page()
905 page_mapcount_reset(page); in reset_page()
906 ClearPageHugeObject(page); in reset_page()
907 page->freelist = NULL; in reset_page()
912 struct page *cursor, *fail; in trylock_zspage()
934 struct page *page, *next; in __free_zspage() local
940 assert_spin_locked(&class->lock); in __free_zspage()
945 next = page = get_first_page(zspage); in __free_zspage()
947 VM_BUG_ON_PAGE(!PageLocked(page), page); in __free_zspage()
948 next = get_next_page(page); in __free_zspage()
949 reset_page(page); in __free_zspage()
950 unlock_page(page); in __free_zspage()
951 dec_zone_page_state(page, NR_ZSPAGES); in __free_zspage()
952 put_page(page); in __free_zspage()
953 page = next; in __free_zspage()
954 } while (page != NULL); in __free_zspage()
958 zs_stat_dec(class, OBJ_ALLOCATED, class->objs_per_zspage); in __free_zspage()
959 atomic_long_sub(class->pages_per_zspage, in __free_zspage()
960 &pool->pages_allocated); in __free_zspage()
967 VM_BUG_ON(list_empty(&zspage->list)); in free_zspage()
983 struct page *page = get_first_page(zspage); in init_zspage() local
985 while (page) { in init_zspage()
986 struct page *next_page; in init_zspage()
990 set_first_obj_offset(page, off); in init_zspage()
992 vaddr = kmap_atomic(page); in init_zspage()
995 while ((off += class->size) < PAGE_SIZE) { in init_zspage()
996 link->next = freeobj++ << OBJ_TAG_BITS; in init_zspage()
997 link += class->size / sizeof(*link); in init_zspage()
1002 * page, which must point to the first object on the next in init_zspage()
1003 * page (if present) in init_zspage()
1005 next_page = get_next_page(page); in init_zspage()
1007 link->next = freeobj++ << OBJ_TAG_BITS; in init_zspage()
1013 link->next = -1UL << OBJ_TAG_BITS; in init_zspage()
1016 page = next_page; in init_zspage()
1024 struct page *pages[]) in create_page_chain()
1027 struct page *page; in create_page_chain() local
1028 struct page *prev_page = NULL; in create_page_chain()
1029 int nr_pages = class->pages_per_zspage; in create_page_chain()
1033 * 1. all pages are linked together using page->freelist in create_page_chain()
1034 * 2. each sub-page point to zspage using page->private in create_page_chain()
1036 * we set PG_private to identify the first page (i.e. no other sub-page in create_page_chain()
1040 page = pages[i]; in create_page_chain()
1041 set_page_private(page, (unsigned long)zspage); in create_page_chain()
1042 page->freelist = NULL; in create_page_chain()
1044 zspage->first_page = page; in create_page_chain()
1045 SetPagePrivate(page); in create_page_chain()
1046 if (unlikely(class->objs_per_zspage == 1 && in create_page_chain()
1047 class->pages_per_zspage == 1)) in create_page_chain()
1048 SetPageHugeObject(page); in create_page_chain()
1050 prev_page->freelist = page; in create_page_chain()
1052 prev_page = page; in create_page_chain()
1057 * Allocate a zspage for the given size class
1064 struct page *pages[ZS_MAX_PAGES_PER_ZSPAGE]; in alloc_zspage()
1071 zspage->magic = ZSPAGE_MAGIC; in alloc_zspage()
1074 for (i = 0; i < class->pages_per_zspage; i++) { in alloc_zspage()
1075 struct page *page; in alloc_zspage() local
1077 page = alloc_page(gfp); in alloc_zspage()
1078 if (!page) { in alloc_zspage()
1079 while (--i >= 0) { in alloc_zspage()
1087 inc_zone_page_state(page, NR_ZSPAGES); in alloc_zspage()
1088 pages[i] = page; in alloc_zspage()
1102 for (i = ZS_ALMOST_FULL; i >= ZS_EMPTY; i--) { in find_get_zspage()
1103 zspage = list_first_entry_or_null(&class->fullness_list[i], in find_get_zspage()
1118 if (area->vm_buf) in __zs_cpu_up()
1120 area->vm_buf = kmalloc(ZS_MAX_ALLOC_SIZE, GFP_KERNEL); in __zs_cpu_up()
1121 if (!area->vm_buf) in __zs_cpu_up()
1122 return -ENOMEM; in __zs_cpu_up()
1128 kfree(area->vm_buf); in __zs_cpu_down()
1129 area->vm_buf = NULL; in __zs_cpu_down()
1133 struct page *pages[2], int off, int size) in __zs_map_object() argument
1137 char *buf = area->vm_buf; in __zs_map_object()
1139 /* disable page faults to match kmap_atomic() return conditions */ in __zs_map_object()
1143 if (area->vm_mm == ZS_MM_WO) in __zs_map_object()
1146 sizes[0] = PAGE_SIZE - off; in __zs_map_object()
1147 sizes[1] = size - sizes[0]; in __zs_map_object()
1149 /* copy object to per-cpu buffer */ in __zs_map_object()
1157 return area->vm_buf; in __zs_map_object()
1161 struct page *pages[2], int off, int size) in __zs_unmap_object() argument
1168 if (area->vm_mm == ZS_MM_RO) in __zs_unmap_object()
1171 buf = area->vm_buf; in __zs_unmap_object()
1173 size -= ZS_HANDLE_SIZE; in __zs_unmap_object()
1176 sizes[0] = PAGE_SIZE - off; in __zs_unmap_object()
1177 sizes[1] = size - sizes[0]; in __zs_unmap_object()
1179 /* copy per-cpu buffer to object */ in __zs_unmap_object()
1188 /* enable page faults to match kunmap_atomic() return conditions */ in __zs_unmap_object()
1212 if (prev->pages_per_zspage == pages_per_zspage && in can_merge()
1213 prev->objs_per_zspage == objs_per_zspage) in can_merge()
1221 return get_zspage_inuse(zspage) == class->objs_per_zspage; in zspage_full()
1226 return atomic_long_read(&pool->pages_allocated); in zs_get_total_pages()
1231 * zs_map_object - get address of allocated object from handle.
1243 * This function returns with preemption and page faults disabled.
1249 struct page *page; in zs_map_object() local
1257 struct page *pages[2]; in zs_map_object()
1261 * Because we use per-cpu mapping areas shared among the in zs_map_object()
1271 obj_to_location(obj, &page, &obj_idx); in zs_map_object()
1272 zspage = get_zspage(page); in zs_map_object()
1278 class = pool->size_class[class_idx]; in zs_map_object()
1279 off = (class->size * obj_idx) & ~PAGE_MASK; in zs_map_object()
1282 area->vm_mm = mm; in zs_map_object()
1283 if (off + class->size <= PAGE_SIZE) { in zs_map_object()
1284 /* this object is contained entirely within a page */ in zs_map_object()
1285 area->vm_addr = kmap_atomic(page); in zs_map_object()
1286 ret = area->vm_addr + off; in zs_map_object()
1291 pages[0] = page; in zs_map_object()
1292 pages[1] = get_next_page(page); in zs_map_object()
1295 ret = __zs_map_object(area, pages, off, class->size); in zs_map_object()
1297 if (likely(!PageHugeObject(page))) in zs_map_object()
1307 struct page *page; in zs_unmap_object() local
1317 obj_to_location(obj, &page, &obj_idx); in zs_unmap_object()
1318 zspage = get_zspage(page); in zs_unmap_object()
1320 class = pool->size_class[class_idx]; in zs_unmap_object()
1321 off = (class->size * obj_idx) & ~PAGE_MASK; in zs_unmap_object()
1324 if (off + class->size <= PAGE_SIZE) in zs_unmap_object()
1325 kunmap_atomic(area->vm_addr); in zs_unmap_object()
1327 struct page *pages[2]; in zs_unmap_object()
1329 pages[0] = page; in zs_unmap_object()
1330 pages[1] = get_next_page(page); in zs_unmap_object()
1333 __zs_unmap_object(area, pages, off, class->size); in zs_unmap_object()
1343 * zs_huge_class_size() - Returns the size (in bytes) of the first huge
1347 * The function returns the size of the first huge class - any object of equal
1348 * or bigger size will be stored in zspage consisting of a single physical
1349 * page.
1353 * Return: the size (in bytes) of the first huge zsmalloc &size_class.
1368 struct page *m_page; in obj_malloc()
1375 offset = obj * class->size; in obj_malloc()
1385 set_freeobj(zspage, link->next >> OBJ_TAG_BITS); in obj_malloc()
1388 link->handle = handle; in obj_malloc()
1390 /* record handle to page->index */ in obj_malloc()
1391 zspage->first_page->index = handle; in obj_malloc()
1404 * zs_malloc - Allocate block of given size from pool.
1406 * @size: size of block to allocate
1411 * Allocation requests with size > ZS_MAX_ALLOC_SIZE will fail.
1413 unsigned long zs_malloc(struct zs_pool *pool, size_t size, gfp_t gfp) in zs_malloc() argument
1420 if (unlikely(!size || size > ZS_MAX_ALLOC_SIZE)) in zs_malloc()
1428 size += ZS_HANDLE_SIZE; in zs_malloc()
1429 class = pool->size_class[get_size_class_index(size)]; in zs_malloc()
1431 spin_lock(&class->lock); in zs_malloc()
1438 spin_unlock(&class->lock); in zs_malloc()
1443 spin_unlock(&class->lock); in zs_malloc()
1451 spin_lock(&class->lock); in zs_malloc()
1455 set_zspage_mapping(zspage, class->index, newfg); in zs_malloc()
1457 atomic_long_add(class->pages_per_zspage, in zs_malloc()
1458 &pool->pages_allocated); in zs_malloc()
1459 zs_stat_inc(class, OBJ_ALLOCATED, class->objs_per_zspage); in zs_malloc()
1463 spin_unlock(&class->lock); in zs_malloc()
1473 struct page *f_page; in obj_free()
1480 f_offset = (class->size * f_objidx) & ~PAGE_MASK; in obj_free()
1487 link->next = get_freeobj(zspage) << OBJ_TAG_BITS; in obj_free()
1490 mod_zspage_inuse(zspage, -1); in obj_free()
1497 struct page *f_page; in zs_free()
1516 class = pool->size_class[class_idx]; in zs_free()
1518 spin_lock(&class->lock); in zs_free()
1533 spin_unlock(&class->lock); in zs_free()
1542 struct page *s_page, *d_page; in zs_object_copy()
1546 int s_size, d_size, size; in zs_object_copy() local
1549 s_size = d_size = class->size; in zs_object_copy()
1554 s_off = (class->size * s_objidx) & ~PAGE_MASK; in zs_object_copy()
1555 d_off = (class->size * d_objidx) & ~PAGE_MASK; in zs_object_copy()
1557 if (s_off + class->size > PAGE_SIZE) in zs_object_copy()
1558 s_size = PAGE_SIZE - s_off; in zs_object_copy()
1560 if (d_off + class->size > PAGE_SIZE) in zs_object_copy()
1561 d_size = PAGE_SIZE - d_off; in zs_object_copy()
1567 size = min(s_size, d_size); in zs_object_copy()
1568 memcpy(d_addr + d_off, s_addr + s_off, size); in zs_object_copy()
1569 written += size; in zs_object_copy()
1571 if (written == class->size) in zs_object_copy()
1574 s_off += size; in zs_object_copy()
1575 s_size -= size; in zs_object_copy()
1576 d_off += size; in zs_object_copy()
1577 d_size -= size; in zs_object_copy()
1585 s_size = class->size - written; in zs_object_copy()
1593 d_size = class->size - written; in zs_object_copy()
1607 struct page *page, int *obj_idx) in find_alloced_obj() argument
1613 void *addr = kmap_atomic(page); in find_alloced_obj()
1615 offset = get_first_obj_offset(page); in find_alloced_obj()
1616 offset += class->size * index; in find_alloced_obj()
1619 head = obj_to_head(page, addr + offset); in find_alloced_obj()
1627 offset += class->size; in find_alloced_obj()
1640 struct page *s_page;
1641 /* Destination page for migration which should be a first page
1643 struct page *d_page;
1654 struct page *s_page = cc->s_page; in migrate_zspage()
1655 struct page *d_page = cc->d_page; in migrate_zspage()
1656 int obj_idx = cc->obj_idx; in migrate_zspage()
1672 ret = -ENOMEM; in migrate_zspage()
1693 cc->s_page = s_page; in migrate_zspage()
1694 cc->obj_idx = obj_idx; in migrate_zspage()
1711 zspage = list_first_entry_or_null(&class->fullness_list[fg[i]], in isolate_zspage()
1724 * putback_zspage - add @zspage into right class's fullness list
1726 * @zspage: target page
1739 set_zspage_mapping(zspage, class->index, fullness); in putback_zspage()
1751 struct page *page = get_first_page(zspage); in lock_zspage() local
1754 lock_page(page); in lock_zspage()
1755 } while ((page = get_next_page(page)) != NULL); in lock_zspage()
1760 return init_pseudo(fc, ZSMALLOC_MAGIC) ? 0 : -ENOMEM; in zs_init_fs_context()
1787 rwlock_init(&zspage->lock); in migrate_lock_init()
1790 static void migrate_read_lock(struct zspage *zspage) __acquires(&zspage->lock) in migrate_read_lock()
1792 read_lock(&zspage->lock); in migrate_read_lock()
1795 static void migrate_read_unlock(struct zspage *zspage) __releases(&zspage->lock) in migrate_read_unlock()
1797 read_unlock(&zspage->lock); in migrate_read_unlock()
1802 write_lock(&zspage->lock); in migrate_write_lock()
1807 write_unlock(&zspage->lock); in migrate_write_unlock()
1810 /* Number of isolated subpage for *page migration* in this zspage */
1813 zspage->isolated++; in inc_zspage_isolation()
1818 zspage->isolated--; in dec_zspage_isolation()
1829 schedule_work(&pool->free_work); in putback_zspage_deferred()
1835 VM_BUG_ON(atomic_long_read(&pool->isolated_pages) <= 0); in zs_pool_dec_isolated()
1836 atomic_long_dec(&pool->isolated_pages); in zs_pool_dec_isolated()
1839 * checks the isolated count under &class->lock after enqueuing in zs_pool_dec_isolated()
1842 if (atomic_long_read(&pool->isolated_pages) == 0 && pool->destroying) in zs_pool_dec_isolated()
1843 wake_up_all(&pool->migration_wait); in zs_pool_dec_isolated()
1847 struct page *newpage, struct page *oldpage) in replace_sub_page()
1849 struct page *page; in replace_sub_page() local
1850 struct page *pages[ZS_MAX_PAGES_PER_ZSPAGE] = {NULL, }; in replace_sub_page()
1853 page = get_first_page(zspage); in replace_sub_page()
1855 if (page == oldpage) in replace_sub_page()
1858 pages[idx] = page; in replace_sub_page()
1860 } while ((page = get_next_page(page)) != NULL); in replace_sub_page()
1865 newpage->index = oldpage->index; in replace_sub_page()
1869 static bool zs_page_isolate(struct page *page, isolate_mode_t mode) in zs_page_isolate() argument
1879 * Page is locked so zspage couldn't be destroyed. For detail, look at in zs_page_isolate()
1882 VM_BUG_ON_PAGE(!PageMovable(page), page); in zs_page_isolate()
1883 VM_BUG_ON_PAGE(PageIsolated(page), page); in zs_page_isolate()
1885 zspage = get_zspage(page); in zs_page_isolate()
1889 * because class_idx is constant unless page is freed so we should get in zs_page_isolate()
1893 mapping = page_mapping(page); in zs_page_isolate()
1894 pool = mapping->private_data; in zs_page_isolate()
1895 class = pool->size_class[class_idx]; in zs_page_isolate()
1897 spin_lock(&class->lock); in zs_page_isolate()
1899 spin_unlock(&class->lock); in zs_page_isolate()
1904 if (list_empty(&zspage->list) && !is_zspage_isolated(zspage)) { in zs_page_isolate()
1905 spin_unlock(&class->lock); in zs_page_isolate()
1913 if (!list_empty(&zspage->list) && !is_zspage_isolated(zspage)) { in zs_page_isolate()
1915 atomic_long_inc(&pool->isolated_pages); in zs_page_isolate()
1920 spin_unlock(&class->lock); in zs_page_isolate()
1925 static int zs_page_migrate(struct address_space *mapping, struct page *newpage, in zs_page_migrate()
1926 struct page *page, enum migrate_mode mode) in zs_page_migrate() argument
1933 struct page *dummy; in zs_page_migrate()
1939 int ret = -EAGAIN; in zs_page_migrate()
1947 return -EINVAL; in zs_page_migrate()
1949 VM_BUG_ON_PAGE(!PageMovable(page), page); in zs_page_migrate()
1950 VM_BUG_ON_PAGE(!PageIsolated(page), page); in zs_page_migrate()
1952 zspage = get_zspage(page); in zs_page_migrate()
1957 pool = mapping->private_data; in zs_page_migrate()
1958 class = pool->size_class[class_idx]; in zs_page_migrate()
1959 offset = get_first_obj_offset(page); in zs_page_migrate()
1961 spin_lock(&class->lock); in zs_page_migrate()
1964 * Set "offset" to end of the page so that every loops in zs_page_migrate()
1971 s_addr = kmap_atomic(page); in zs_page_migrate()
1973 head = obj_to_head(page, s_addr + pos); in zs_page_migrate()
1979 pos += class->size; in zs_page_migrate()
1990 addr += class->size) { in zs_page_migrate()
1991 head = obj_to_head(page, addr); in zs_page_migrate()
2006 replace_sub_page(class, zspage, newpage, page); in zs_page_migrate()
2012 * Page migration is done so let's putback isolated zspage to in zs_page_migrate()
2013 * the list if @page is final isolated subpage in the zspage. in zs_page_migrate()
2019 * Also, we ensure that everyone can see pool->destroying before in zs_page_migrate()
2026 if (page_zone(newpage) != page_zone(page)) { in zs_page_migrate()
2027 dec_zone_page_state(page, NR_ZSPAGES); in zs_page_migrate()
2031 reset_page(page); in zs_page_migrate()
2032 put_page(page); in zs_page_migrate()
2033 page = newpage; in zs_page_migrate()
2038 addr += class->size) { in zs_page_migrate()
2039 head = obj_to_head(page, addr); in zs_page_migrate()
2048 spin_unlock(&class->lock); in zs_page_migrate()
2054 static void zs_page_putback(struct page *page) in zs_page_putback() argument
2063 VM_BUG_ON_PAGE(!PageMovable(page), page); in zs_page_putback()
2064 VM_BUG_ON_PAGE(!PageIsolated(page), page); in zs_page_putback()
2066 zspage = get_zspage(page); in zs_page_putback()
2068 mapping = page_mapping(page); in zs_page_putback()
2069 pool = mapping->private_data; in zs_page_putback()
2070 class = pool->size_class[class_idx]; in zs_page_putback()
2072 spin_lock(&class->lock); in zs_page_putback()
2082 spin_unlock(&class->lock); in zs_page_putback()
2093 pool->inode = alloc_anon_inode(zsmalloc_mnt->mnt_sb); in zs_register_migration()
2094 if (IS_ERR(pool->inode)) { in zs_register_migration()
2095 pool->inode = NULL; in zs_register_migration()
2099 pool->inode->i_mapping->private_data = pool; in zs_register_migration()
2100 pool->inode->i_mapping->a_ops = &zsmalloc_aops; in zs_register_migration()
2106 return atomic_long_read(&pool->isolated_pages) == 0; in pool_isolated_are_drained()
2119 wait_event(pool->migration_wait, in wait_for_isolated_drain()
2125 pool->destroying = true; in zs_unregister_migration()
2128 * pool->destroying. Thus pool->isolated pages will either be 0 in which in zs_unregister_migration()
2129 * case we don't care, or it will be > 0 and pool->destroying will in zs_unregister_migration()
2134 flush_work(&pool->free_work); in zs_unregister_migration()
2135 iput(pool->inode); in zs_unregister_migration()
2154 class = pool->size_class[i]; in async_free_zspage()
2155 if (class->index != i) in async_free_zspage()
2158 spin_lock(&class->lock); in async_free_zspage()
2159 list_splice_init(&class->fullness_list[ZS_EMPTY], &free_pages); in async_free_zspage()
2160 spin_unlock(&class->lock); in async_free_zspage()
2165 list_del(&zspage->list); in async_free_zspage()
2170 class = pool->size_class[class_idx]; in async_free_zspage()
2171 spin_lock(&class->lock); in async_free_zspage()
2172 __free_zspage(pool, pool->size_class[class_idx], zspage); in async_free_zspage()
2173 spin_unlock(&class->lock); in async_free_zspage()
2179 schedule_work(&pool->free_work); in kick_deferred_free()
2184 INIT_WORK(&pool->free_work, async_free_zspage); in init_deferred_free()
2189 struct page *page = get_first_page(zspage); in SetZsPageMovable() local
2192 WARN_ON(!trylock_page(page)); in SetZsPageMovable()
2193 __SetPageMovable(page, pool->inode->i_mapping); in SetZsPageMovable()
2194 unlock_page(page); in SetZsPageMovable()
2195 } while ((page = get_next_page(page)) != NULL); in SetZsPageMovable()
2213 obj_wasted = obj_allocated - obj_used; in zs_can_compact()
2214 obj_wasted /= class->objs_per_zspage; in zs_can_compact()
2216 return obj_wasted * class->pages_per_zspage; in zs_can_compact()
2225 spin_lock(&class->lock); in __zs_compact()
2253 pool->stats.pages_compacted += class->pages_per_zspage; in __zs_compact()
2255 spin_unlock(&class->lock); in __zs_compact()
2257 spin_lock(&class->lock); in __zs_compact()
2263 spin_unlock(&class->lock); in __zs_compact()
2271 for (i = ZS_SIZE_CLASSES - 1; i >= 0; i--) { in zs_compact()
2272 class = pool->size_class[i]; in zs_compact()
2275 if (class->index != i) in zs_compact()
2280 return pool->stats.pages_compacted; in zs_compact()
2286 memcpy(stats, &pool->stats, sizeof(struct zs_pool_stats)); in zs_pool_stats()
2297 pages_freed = pool->stats.pages_compacted; in zs_shrinker_scan()
2303 pages_freed = zs_compact(pool) - pages_freed; in zs_shrinker_scan()
2317 for (i = ZS_SIZE_CLASSES - 1; i >= 0; i--) { in zs_shrinker_count()
2318 class = pool->size_class[i]; in zs_shrinker_count()
2321 if (class->index != i) in zs_shrinker_count()
2332 unregister_shrinker(&pool->shrinker); in zs_unregister_shrinker()
2337 pool->shrinker.scan_objects = zs_shrinker_scan; in zs_register_shrinker()
2338 pool->shrinker.count_objects = zs_shrinker_count; in zs_register_shrinker()
2339 pool->shrinker.batch = 0; in zs_register_shrinker()
2340 pool->shrinker.seeks = DEFAULT_SEEKS; in zs_register_shrinker()
2342 return register_shrinker(&pool->shrinker); in zs_register_shrinker()
2346 * zs_create_pool - Creates an allocation pool to work from.
2367 pool->name = kstrdup(name, GFP_KERNEL); in zs_create_pool()
2368 if (!pool->name) in zs_create_pool()
2372 init_waitqueue_head(&pool->migration_wait); in zs_create_pool()
2379 * Iterate reversely, because, size of size_class that we want to use in zs_create_pool()
2380 * for merging should be larger or equal to current size. in zs_create_pool()
2382 for (i = ZS_SIZE_CLASSES - 1; i >= 0; i--) { in zs_create_pool()
2383 int size; in zs_create_pool() local
2389 size = ZS_MIN_ALLOC_SIZE + i * ZS_SIZE_CLASS_DELTA; in zs_create_pool()
2390 if (size > ZS_MAX_ALLOC_SIZE) in zs_create_pool()
2391 size = ZS_MAX_ALLOC_SIZE; in zs_create_pool()
2392 pages_per_zspage = get_pages_per_zspage(size); in zs_create_pool()
2393 objs_per_zspage = pages_per_zspage * PAGE_SIZE / size; in zs_create_pool()
2397 * so huge_class_size holds the size of the first huge in zs_create_pool()
2403 huge_class_size = size; in zs_create_pool()
2407 * unconditionally adds handle size before it performs in zs_create_pool()
2408 * size class search - so object may be smaller than in zs_create_pool()
2409 * huge class size, yet it still can end up in the huge in zs_create_pool()
2413 huge_class_size -= (ZS_HANDLE_SIZE - 1); in zs_create_pool()
2418 * as alloc/free for that size. Although it is natural that we in zs_create_pool()
2419 * have one size_class for each size, there is a chance that we in zs_create_pool()
2427 pool->size_class[i] = prev_class; in zs_create_pool()
2436 class->size = size; in zs_create_pool()
2437 class->index = i; in zs_create_pool()
2438 class->pages_per_zspage = pages_per_zspage; in zs_create_pool()
2439 class->objs_per_zspage = objs_per_zspage; in zs_create_pool()
2440 spin_lock_init(&class->lock); in zs_create_pool()
2441 pool->size_class[i] = class; in zs_create_pool()
2444 INIT_LIST_HEAD(&class->fullness_list[fullness]); in zs_create_pool()
2481 struct size_class *class = pool->size_class[i]; in zs_destroy_pool()
2486 if (class->index != i) in zs_destroy_pool()
2490 if (!list_empty(&class->fullness_list[fg])) { in zs_destroy_pool()
2491 pr_info("Freeing non-empty class with size %db, fullness group %d\n", in zs_destroy_pool()
2492 class->size, fg); in zs_destroy_pool()
2499 kfree(pool->name); in zs_destroy_pool()