diff options
Diffstat (limited to 'mm/slub.c')
-rw-r--r-- | mm/slub.c | 132 |
1 files changed, 81 insertions, 51 deletions
@@ -281,11 +281,40 @@ static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |||
281 | return (p - addr) / s->size; | 281 | return (p - addr) / s->size; |
282 | } | 282 | } |
283 | 283 | ||
284 | static inline size_t slab_ksize(const struct kmem_cache *s) | ||
285 | { | ||
286 | #ifdef CONFIG_SLUB_DEBUG | ||
287 | /* | ||
288 | * Debugging requires use of the padding between object | ||
289 | * and whatever may come after it. | ||
290 | */ | ||
291 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | ||
292 | return s->objsize; | ||
293 | |||
294 | #endif | ||
295 | /* | ||
296 | * If we have the need to store the freelist pointer | ||
297 | * back there or track user information then we can | ||
298 | * only use the space before that information. | ||
299 | */ | ||
300 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | ||
301 | return s->inuse; | ||
302 | /* | ||
303 | * Else we can use all the padding etc for the allocation | ||
304 | */ | ||
305 | return s->size; | ||
306 | } | ||
307 | |||
308 | static inline int order_objects(int order, unsigned long size, int reserved) | ||
309 | { | ||
310 | return ((PAGE_SIZE << order) - reserved) / size; | ||
311 | } | ||
312 | |||
284 | static inline struct kmem_cache_order_objects oo_make(int order, | 313 | static inline struct kmem_cache_order_objects oo_make(int order, |
285 | unsigned long size) | 314 | unsigned long size, int reserved) |
286 | { | 315 | { |
287 | struct kmem_cache_order_objects x = { | 316 | struct kmem_cache_order_objects x = { |
288 | (order << OO_SHIFT) + (PAGE_SIZE << order) / size | 317 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
289 | }; | 318 | }; |
290 | 319 | ||
291 | return x; | 320 | return x; |
@@ -617,7 +646,7 @@ static int slab_pad_check(struct kmem_cache *s, struct page *page) | |||
617 | return 1; | 646 | return 1; |
618 | 647 | ||
619 | start = page_address(page); | 648 | start = page_address(page); |
620 | length = (PAGE_SIZE << compound_order(page)); | 649 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
621 | end = start + length; | 650 | end = start + length; |
622 | remainder = length % s->size; | 651 | remainder = length % s->size; |
623 | if (!remainder) | 652 | if (!remainder) |
@@ -698,7 +727,7 @@ static int check_slab(struct kmem_cache *s, struct page *page) | |||
698 | return 0; | 727 | return 0; |
699 | } | 728 | } |
700 | 729 | ||
701 | maxobj = (PAGE_SIZE << compound_order(page)) / s->size; | 730 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
702 | if (page->objects > maxobj) { | 731 | if (page->objects > maxobj) { |
703 | slab_err(s, page, "objects %u > max %u", | 732 | slab_err(s, page, "objects %u > max %u", |
704 | s->name, page->objects, maxobj); | 733 | s->name, page->objects, maxobj); |
@@ -748,7 +777,7 @@ static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |||
748 | nr++; | 777 | nr++; |
749 | } | 778 | } |
750 | 779 | ||
751 | max_objects = (PAGE_SIZE << compound_order(page)) / s->size; | 780 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
752 | if (max_objects > MAX_OBJS_PER_PAGE) | 781 | if (max_objects > MAX_OBJS_PER_PAGE) |
753 | max_objects = MAX_OBJS_PER_PAGE; | 782 | max_objects = MAX_OBJS_PER_PAGE; |
754 | 783 | ||
@@ -800,7 +829,7 @@ static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) | |||
800 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object) | 829 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, void *object) |
801 | { | 830 | { |
802 | flags &= gfp_allowed_mask; | 831 | flags &= gfp_allowed_mask; |
803 | kmemcheck_slab_alloc(s, flags, object, s->objsize); | 832 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); |
804 | kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags); | 833 | kmemleak_alloc_recursive(object, s->objsize, 1, s->flags, flags); |
805 | } | 834 | } |
806 | 835 | ||
@@ -1249,21 +1278,38 @@ static void __free_slab(struct kmem_cache *s, struct page *page) | |||
1249 | __free_pages(page, order); | 1278 | __free_pages(page, order); |
1250 | } | 1279 | } |
1251 | 1280 | ||
1281 | #define need_reserve_slab_rcu \ | ||
1282 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | ||
1283 | |||
1252 | static void rcu_free_slab(struct rcu_head *h) | 1284 | static void rcu_free_slab(struct rcu_head *h) |
1253 | { | 1285 | { |
1254 | struct page *page; | 1286 | struct page *page; |
1255 | 1287 | ||
1256 | page = container_of((struct list_head *)h, struct page, lru); | 1288 | if (need_reserve_slab_rcu) |
1289 | page = virt_to_head_page(h); | ||
1290 | else | ||
1291 | page = container_of((struct list_head *)h, struct page, lru); | ||
1292 | |||
1257 | __free_slab(page->slab, page); | 1293 | __free_slab(page->slab, page); |
1258 | } | 1294 | } |
1259 | 1295 | ||
1260 | static void free_slab(struct kmem_cache *s, struct page *page) | 1296 | static void free_slab(struct kmem_cache *s, struct page *page) |
1261 | { | 1297 | { |
1262 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | 1298 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { |
1263 | /* | 1299 | struct rcu_head *head; |
1264 | * RCU free overloads the RCU head over the LRU | 1300 | |
1265 | */ | 1301 | if (need_reserve_slab_rcu) { |
1266 | struct rcu_head *head = (void *)&page->lru; | 1302 | int order = compound_order(page); |
1303 | int offset = (PAGE_SIZE << order) - s->reserved; | ||
1304 | |||
1305 | VM_BUG_ON(s->reserved != sizeof(*head)); | ||
1306 | head = page_address(page) + offset; | ||
1307 | } else { | ||
1308 | /* | ||
1309 | * RCU free overloads the RCU head over the LRU | ||
1310 | */ | ||
1311 | head = (void *)&page->lru; | ||
1312 | } | ||
1267 | 1313 | ||
1268 | call_rcu(head, rcu_free_slab); | 1314 | call_rcu(head, rcu_free_slab); |
1269 | } else | 1315 | } else |
@@ -1988,13 +2034,13 @@ static int slub_nomerge; | |||
1988 | * the smallest order which will fit the object. | 2034 | * the smallest order which will fit the object. |
1989 | */ | 2035 | */ |
1990 | static inline int slab_order(int size, int min_objects, | 2036 | static inline int slab_order(int size, int min_objects, |
1991 | int max_order, int fract_leftover) | 2037 | int max_order, int fract_leftover, int reserved) |
1992 | { | 2038 | { |
1993 | int order; | 2039 | int order; |
1994 | int rem; | 2040 | int rem; |
1995 | int min_order = slub_min_order; | 2041 | int min_order = slub_min_order; |
1996 | 2042 | ||
1997 | if ((PAGE_SIZE << min_order) / size > MAX_OBJS_PER_PAGE) | 2043 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
1998 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; | 2044 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
1999 | 2045 | ||
2000 | for (order = max(min_order, | 2046 | for (order = max(min_order, |
@@ -2003,10 +2049,10 @@ static inline int slab_order(int size, int min_objects, | |||
2003 | 2049 | ||
2004 | unsigned long slab_size = PAGE_SIZE << order; | 2050 | unsigned long slab_size = PAGE_SIZE << order; |
2005 | 2051 | ||
2006 | if (slab_size < min_objects * size) | 2052 | if (slab_size < min_objects * size + reserved) |
2007 | continue; | 2053 | continue; |
2008 | 2054 | ||
2009 | rem = slab_size % size; | 2055 | rem = (slab_size - reserved) % size; |
2010 | 2056 | ||
2011 | if (rem <= slab_size / fract_leftover) | 2057 | if (rem <= slab_size / fract_leftover) |
2012 | break; | 2058 | break; |
@@ -2016,7 +2062,7 @@ static inline int slab_order(int size, int min_objects, | |||
2016 | return order; | 2062 | return order; |
2017 | } | 2063 | } |
2018 | 2064 | ||
2019 | static inline int calculate_order(int size) | 2065 | static inline int calculate_order(int size, int reserved) |
2020 | { | 2066 | { |
2021 | int order; | 2067 | int order; |
2022 | int min_objects; | 2068 | int min_objects; |
@@ -2034,14 +2080,14 @@ static inline int calculate_order(int size) | |||
2034 | min_objects = slub_min_objects; | 2080 | min_objects = slub_min_objects; |
2035 | if (!min_objects) | 2081 | if (!min_objects) |
2036 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | 2082 | min_objects = 4 * (fls(nr_cpu_ids) + 1); |
2037 | max_objects = (PAGE_SIZE << slub_max_order)/size; | 2083 | max_objects = order_objects(slub_max_order, size, reserved); |
2038 | min_objects = min(min_objects, max_objects); | 2084 | min_objects = min(min_objects, max_objects); |
2039 | 2085 | ||
2040 | while (min_objects > 1) { | 2086 | while (min_objects > 1) { |
2041 | fraction = 16; | 2087 | fraction = 16; |
2042 | while (fraction >= 4) { | 2088 | while (fraction >= 4) { |
2043 | order = slab_order(size, min_objects, | 2089 | order = slab_order(size, min_objects, |
2044 | slub_max_order, fraction); | 2090 | slub_max_order, fraction, reserved); |
2045 | if (order <= slub_max_order) | 2091 | if (order <= slub_max_order) |
2046 | return order; | 2092 | return order; |
2047 | fraction /= 2; | 2093 | fraction /= 2; |
@@ -2053,14 +2099,14 @@ static inline int calculate_order(int size) | |||
2053 | * We were unable to place multiple objects in a slab. Now | 2099 | * We were unable to place multiple objects in a slab. Now |
2054 | * lets see if we can place a single object there. | 2100 | * lets see if we can place a single object there. |
2055 | */ | 2101 | */ |
2056 | order = slab_order(size, 1, slub_max_order, 1); | 2102 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
2057 | if (order <= slub_max_order) | 2103 | if (order <= slub_max_order) |
2058 | return order; | 2104 | return order; |
2059 | 2105 | ||
2060 | /* | 2106 | /* |
2061 | * Doh this slab cannot be placed using slub_max_order. | 2107 | * Doh this slab cannot be placed using slub_max_order. |
2062 | */ | 2108 | */ |
2063 | order = slab_order(size, 1, MAX_ORDER, 1); | 2109 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
2064 | if (order < MAX_ORDER) | 2110 | if (order < MAX_ORDER) |
2065 | return order; | 2111 | return order; |
2066 | return -ENOSYS; | 2112 | return -ENOSYS; |
@@ -2311,7 +2357,7 @@ static int calculate_sizes(struct kmem_cache *s, int forced_order) | |||
2311 | if (forced_order >= 0) | 2357 | if (forced_order >= 0) |
2312 | order = forced_order; | 2358 | order = forced_order; |
2313 | else | 2359 | else |
2314 | order = calculate_order(size); | 2360 | order = calculate_order(size, s->reserved); |
2315 | 2361 | ||
2316 | if (order < 0) | 2362 | if (order < 0) |
2317 | return 0; | 2363 | return 0; |
@@ -2329,8 +2375,8 @@ static int calculate_sizes(struct kmem_cache *s, int forced_order) | |||
2329 | /* | 2375 | /* |
2330 | * Determine the number of objects per slab | 2376 | * Determine the number of objects per slab |
2331 | */ | 2377 | */ |
2332 | s->oo = oo_make(order, size); | 2378 | s->oo = oo_make(order, size, s->reserved); |
2333 | s->min = oo_make(get_order(size), size); | 2379 | s->min = oo_make(get_order(size), size, s->reserved); |
2334 | if (oo_objects(s->oo) > oo_objects(s->max)) | 2380 | if (oo_objects(s->oo) > oo_objects(s->max)) |
2335 | s->max = s->oo; | 2381 | s->max = s->oo; |
2336 | 2382 | ||
@@ -2349,6 +2395,10 @@ static int kmem_cache_open(struct kmem_cache *s, | |||
2349 | s->objsize = size; | 2395 | s->objsize = size; |
2350 | s->align = align; | 2396 | s->align = align; |
2351 | s->flags = kmem_cache_flags(size, flags, name, ctor); | 2397 | s->flags = kmem_cache_flags(size, flags, name, ctor); |
2398 | s->reserved = 0; | ||
2399 | |||
2400 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) | ||
2401 | s->reserved = sizeof(struct rcu_head); | ||
2352 | 2402 | ||
2353 | if (!calculate_sizes(s, -1)) | 2403 | if (!calculate_sizes(s, -1)) |
2354 | goto error; | 2404 | goto error; |
@@ -2399,12 +2449,6 @@ unsigned int kmem_cache_size(struct kmem_cache *s) | |||
2399 | } | 2449 | } |
2400 | EXPORT_SYMBOL(kmem_cache_size); | 2450 | EXPORT_SYMBOL(kmem_cache_size); |
2401 | 2451 | ||
2402 | const char *kmem_cache_name(struct kmem_cache *s) | ||
2403 | { | ||
2404 | return s->name; | ||
2405 | } | ||
2406 | EXPORT_SYMBOL(kmem_cache_name); | ||
2407 | |||
2408 | static void list_slab_objects(struct kmem_cache *s, struct page *page, | 2452 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
2409 | const char *text) | 2453 | const char *text) |
2410 | { | 2454 | { |
@@ -2696,7 +2740,6 @@ EXPORT_SYMBOL(__kmalloc_node); | |||
2696 | size_t ksize(const void *object) | 2740 | size_t ksize(const void *object) |
2697 | { | 2741 | { |
2698 | struct page *page; | 2742 | struct page *page; |
2699 | struct kmem_cache *s; | ||
2700 | 2743 | ||
2701 | if (unlikely(object == ZERO_SIZE_PTR)) | 2744 | if (unlikely(object == ZERO_SIZE_PTR)) |
2702 | return 0; | 2745 | return 0; |
@@ -2707,28 +2750,8 @@ size_t ksize(const void *object) | |||
2707 | WARN_ON(!PageCompound(page)); | 2750 | WARN_ON(!PageCompound(page)); |
2708 | return PAGE_SIZE << compound_order(page); | 2751 | return PAGE_SIZE << compound_order(page); |
2709 | } | 2752 | } |
2710 | s = page->slab; | ||
2711 | |||
2712 | #ifdef CONFIG_SLUB_DEBUG | ||
2713 | /* | ||
2714 | * Debugging requires use of the padding between object | ||
2715 | * and whatever may come after it. | ||
2716 | */ | ||
2717 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | ||
2718 | return s->objsize; | ||
2719 | 2753 | ||
2720 | #endif | 2754 | return slab_ksize(page->slab); |
2721 | /* | ||
2722 | * If we have the need to store the freelist pointer | ||
2723 | * back there or track user information then we can | ||
2724 | * only use the space before that information. | ||
2725 | */ | ||
2726 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | ||
2727 | return s->inuse; | ||
2728 | /* | ||
2729 | * Else we can use all the padding etc for the allocation | ||
2730 | */ | ||
2731 | return s->size; | ||
2732 | } | 2755 | } |
2733 | EXPORT_SYMBOL(ksize); | 2756 | EXPORT_SYMBOL(ksize); |
2734 | 2757 | ||
@@ -4017,6 +4040,12 @@ static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |||
4017 | } | 4040 | } |
4018 | SLAB_ATTR_RO(destroy_by_rcu); | 4041 | SLAB_ATTR_RO(destroy_by_rcu); |
4019 | 4042 | ||
4043 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) | ||
4044 | { | ||
4045 | return sprintf(buf, "%d\n", s->reserved); | ||
4046 | } | ||
4047 | SLAB_ATTR_RO(reserved); | ||
4048 | |||
4020 | #ifdef CONFIG_SLUB_DEBUG | 4049 | #ifdef CONFIG_SLUB_DEBUG |
4021 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) | 4050 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4022 | { | 4051 | { |
@@ -4303,6 +4332,7 @@ static struct attribute *slab_attrs[] = { | |||
4303 | &reclaim_account_attr.attr, | 4332 | &reclaim_account_attr.attr, |
4304 | &destroy_by_rcu_attr.attr, | 4333 | &destroy_by_rcu_attr.attr, |
4305 | &shrink_attr.attr, | 4334 | &shrink_attr.attr, |
4335 | &reserved_attr.attr, | ||
4306 | #ifdef CONFIG_SLUB_DEBUG | 4336 | #ifdef CONFIG_SLUB_DEBUG |
4307 | &total_objects_attr.attr, | 4337 | &total_objects_attr.attr, |
4308 | &slabs_attr.attr, | 4338 | &slabs_attr.attr, |