blob: 45ca123e800219777001f75113324f6cfd60486f [file] [log] [blame]
Christoph Lameter81819f02007-05-06 14:49:36 -07001#ifndef _LINUX_SLUB_DEF_H
2#define _LINUX_SLUB_DEF_H
3
4/*
5 * SLUB : A Slab allocator without object queues.
6 *
Christoph Lametercde53532008-07-04 09:59:22 -07007 * (C) 2007 SGI, Christoph Lameter
Christoph Lameter81819f02007-05-06 14:49:36 -07008 */
9#include <linux/types.h>
10#include <linux/gfp.h>
11#include <linux/workqueue.h>
12#include <linux/kobject.h>
13
Richard Kennedy4a923792010-10-21 10:29:19 +010014#include <linux/kmemleak.h>
Li Zefan039ca4e2010-05-26 17:22:17 +080015
Christoph Lameter8ff12cf2008-02-07 17:47:41 -080016enum stat_item {
17 ALLOC_FASTPATH, /* Allocation from cpu slab */
18 ALLOC_SLOWPATH, /* Allocation by getting a new cpu slab */
19 FREE_FASTPATH, /* Free to cpu slub */
20 FREE_SLOWPATH, /* Freeing not to cpu slab */
21 FREE_FROZEN, /* Freeing to frozen slab */
22 FREE_ADD_PARTIAL, /* Freeing moves slab to partial list */
23 FREE_REMOVE_PARTIAL, /* Freeing removes last object */
24 ALLOC_FROM_PARTIAL, /* Cpu slab acquired from partial list */
25 ALLOC_SLAB, /* Cpu slab acquired from page allocator */
26 ALLOC_REFILL, /* Refill cpu slab from slab freelist */
27 FREE_SLAB, /* Slab freed to the page allocator */
28 CPUSLAB_FLUSH, /* Abandoning of the cpu slab */
29 DEACTIVATE_FULL, /* Cpu slab was full when deactivated */
30 DEACTIVATE_EMPTY, /* Cpu slab was empty when deactivated */
31 DEACTIVATE_TO_HEAD, /* Cpu slab was moved to the head of partials */
32 DEACTIVATE_TO_TAIL, /* Cpu slab was moved to the tail of partials */
33 DEACTIVATE_REMOTE_FREES,/* Slab contained remotely freed objects */
Christoph Lameter65c33762008-04-14 19:11:40 +030034 ORDER_FALLBACK, /* Number of times fallback was necessary */
Christoph Lameter4fdccdf2011-03-22 13:35:00 -050035 CMPXCHG_DOUBLE_CPU_FAIL,/* Failure of this_cpu_cmpxchg_double */
Christoph Lameter8ff12cf2008-02-07 17:47:41 -080036 NR_SLUB_STAT_ITEMS };
37
Christoph Lameterdfb4f092007-10-16 01:26:05 -070038struct kmem_cache_cpu {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -060039 void **freelist; /* Pointer to next available object */
40#ifdef CONFIG_CMPXCHG_LOCAL
41 unsigned long tid; /* Globally unique transaction id */
42#endif
Christoph Lameterda89b792008-01-07 23:20:31 -080043 struct page *page; /* The slab from which we are allocating */
44 int node; /* The node of the page (or -1 for debug) */
Christoph Lameter8ff12cf2008-02-07 17:47:41 -080045#ifdef CONFIG_SLUB_STATS
46 unsigned stat[NR_SLUB_STAT_ITEMS];
47#endif
Christoph Lameter4c93c3552007-10-16 01:26:08 -070048};
Christoph Lameterdfb4f092007-10-16 01:26:05 -070049
Christoph Lameter81819f02007-05-06 14:49:36 -070050struct kmem_cache_node {
51 spinlock_t list_lock; /* Protect partial list and nr_partial */
52 unsigned long nr_partial;
Christoph Lameter81819f02007-05-06 14:49:36 -070053 struct list_head partial;
Christoph Lameter0c710012007-07-17 04:03:24 -070054#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +030055 atomic_long_t nr_slabs;
Christoph Lameter205ab992008-04-14 19:11:40 +030056 atomic_long_t total_objects;
Christoph Lameter643b1132007-05-06 14:49:42 -070057 struct list_head full;
Christoph Lameter0c710012007-07-17 04:03:24 -070058#endif
Christoph Lameter81819f02007-05-06 14:49:36 -070059};
60
61/*
Christoph Lameter834f3d12008-04-14 19:11:31 +030062 * Word size structure that can be atomically updated or read and that
63 * contains both the order and the number of objects that a slab of the
64 * given order would contain.
65 */
66struct kmem_cache_order_objects {
67 unsigned long x;
68};
69
70/*
Christoph Lameter81819f02007-05-06 14:49:36 -070071 * Slab cache management.
72 */
73struct kmem_cache {
Namhyung Kim1b5ad242010-08-07 14:29:22 +020074 struct kmem_cache_cpu __percpu *cpu_slab;
Christoph Lameter81819f02007-05-06 14:49:36 -070075 /* Used for retriving partial slabs etc */
76 unsigned long flags;
Christoph Lameter1a757fe2011-02-25 11:38:51 -060077 unsigned long min_partial;
Christoph Lameter81819f02007-05-06 14:49:36 -070078 int size; /* The size of an object including meta data */
79 int objsize; /* The size of an object without meta data */
80 int offset; /* Free pointer offset. */
Christoph Lameter834f3d12008-04-14 19:11:31 +030081 struct kmem_cache_order_objects oo;
Christoph Lameter81819f02007-05-06 14:49:36 -070082
Christoph Lameter81819f02007-05-06 14:49:36 -070083 /* Allocation and freeing of slabs */
Christoph Lameter205ab992008-04-14 19:11:40 +030084 struct kmem_cache_order_objects max;
Christoph Lameter65c33762008-04-14 19:11:40 +030085 struct kmem_cache_order_objects min;
Christoph Lameterb7a49f02008-02-14 14:21:32 -080086 gfp_t allocflags; /* gfp flags to use on each alloc */
Christoph Lameter81819f02007-05-06 14:49:36 -070087 int refcount; /* Refcount for slab cache destroy */
Alexey Dobriyan51cc5062008-07-25 19:45:34 -070088 void (*ctor)(void *);
Christoph Lameter81819f02007-05-06 14:49:36 -070089 int inuse; /* Offset to metadata */
90 int align; /* Alignment */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +080091 int reserved; /* Reserved bytes at the end of slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -070092 const char *name; /* Name (only for display!) */
93 struct list_head list; /* List of slab caches */
Christoph Lameterab4d5ed2010-10-05 13:57:26 -050094#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -070095 struct kobject kobj; /* For sysfs */
Christoph Lameter0c710012007-07-17 04:03:24 -070096#endif
Christoph Lameter81819f02007-05-06 14:49:36 -070097
98#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -080099 /*
100 * Defragmentation by allocating from a remote node.
101 */
102 int remote_node_defrag_ratio;
Christoph Lameter81819f02007-05-06 14:49:36 -0700103#endif
Christoph Lameter7340cc82010-09-28 08:10:26 -0500104 struct kmem_cache_node *node[MAX_NUMNODES];
Christoph Lameter81819f02007-05-06 14:49:36 -0700105};
106
107/*
108 * Kmalloc subsystem.
109 */
FUJITA Tomonoria6eb9fe2010-08-10 18:03:22 -0700110#if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
111#define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
Christoph Lameter4b356be2007-06-16 10:16:13 -0700112#else
113#define KMALLOC_MIN_SIZE 8
114#endif
115
116#define KMALLOC_SHIFT_LOW ilog2(KMALLOC_MIN_SIZE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700117
FUJITA Tomonoria6eb9fe2010-08-10 18:03:22 -0700118#ifdef ARCH_DMA_MINALIGN
119#define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
120#else
David Woodhouse4581ced2010-05-19 12:02:14 +0100121#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
122#endif
123
124#ifndef ARCH_SLAB_MINALIGN
125#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
126#endif
127
Christoph Lameter81819f02007-05-06 14:49:36 -0700128/*
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500129 * Maximum kmalloc object size handled by SLUB. Larger object allocations
130 * are passed through to the page allocator. The page allocator "fastpath"
131 * is relatively slow so we need this value sufficiently high so that
132 * performance critical objects are allocated through the SLUB fastpath.
133 *
134 * This should be dropped to PAGE_SIZE / 2 once the page allocator
135 * "fastpath" becomes competitive with the slab allocator fastpaths.
136 */
Pekka Enberg51735a72009-02-20 12:21:33 +0200137#define SLUB_MAX_SIZE (2 * PAGE_SIZE)
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500138
Pekka Enberg51735a72009-02-20 12:21:33 +0200139#define SLUB_PAGE_SHIFT (PAGE_SHIFT + 2)
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500140
Christoph Lameter756dee72009-12-18 16:26:21 -0600141#ifdef CONFIG_ZONE_DMA
142#define SLUB_DMA __GFP_DMA
Christoph Lameter756dee72009-12-18 16:26:21 -0600143#else
144/* Disable DMA functionality */
145#define SLUB_DMA (__force gfp_t)0
Christoph Lameter756dee72009-12-18 16:26:21 -0600146#endif
147
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500148/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700149 * We keep the general caches in an array of slab caches that are used for
150 * 2^x bytes of allocations.
151 */
Christoph Lameter51df1142010-08-20 12:37:15 -0500152extern struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
Christoph Lameter81819f02007-05-06 14:49:36 -0700153
154/*
155 * Sorry that the following has to be that ugly but some versions of GCC
156 * have trouble with constant propagation and loops.
157 */
Christoph Lameteraa137f92007-08-31 00:48:45 -0700158static __always_inline int kmalloc_index(size_t size)
Christoph Lameter81819f02007-05-06 14:49:36 -0700159{
Christoph Lameter272c1d22007-06-08 13:46:49 -0700160 if (!size)
161 return 0;
Christoph Lameter614410d2007-05-06 14:49:38 -0700162
Christoph Lameter4b356be2007-06-16 10:16:13 -0700163 if (size <= KMALLOC_MIN_SIZE)
164 return KMALLOC_SHIFT_LOW;
165
Aaro Koskinenacdfcd02009-08-28 14:28:54 +0300166 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
Christoph Lameter81819f02007-05-06 14:49:36 -0700167 return 1;
Aaro Koskinenacdfcd02009-08-28 14:28:54 +0300168 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
Christoph Lameter81819f02007-05-06 14:49:36 -0700169 return 2;
170 if (size <= 8) return 3;
171 if (size <= 16) return 4;
172 if (size <= 32) return 5;
173 if (size <= 64) return 6;
174 if (size <= 128) return 7;
175 if (size <= 256) return 8;
176 if (size <= 512) return 9;
177 if (size <= 1024) return 10;
178 if (size <= 2 * 1024) return 11;
Christoph Lameter6446faa2008-02-15 23:45:26 -0800179 if (size <= 4 * 1024) return 12;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700180/*
181 * The following is only needed to support architectures with a larger page
182 * size than 4k.
183 */
Christoph Lameter81819f02007-05-06 14:49:36 -0700184 if (size <= 8 * 1024) return 13;
185 if (size <= 16 * 1024) return 14;
186 if (size <= 32 * 1024) return 15;
187 if (size <= 64 * 1024) return 16;
188 if (size <= 128 * 1024) return 17;
189 if (size <= 256 * 1024) return 18;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700190 if (size <= 512 * 1024) return 19;
Christoph Lameter81819f02007-05-06 14:49:36 -0700191 if (size <= 1024 * 1024) return 20;
Christoph Lameter81819f02007-05-06 14:49:36 -0700192 if (size <= 2 * 1024 * 1024) return 21;
Christoph Lameter81819f02007-05-06 14:49:36 -0700193 return -1;
194
195/*
196 * What we really wanted to do and cannot do because of compiler issues is:
197 * int i;
198 * for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++)
199 * if (size <= (1 << i))
200 * return i;
201 */
202}
203
204/*
205 * Find the slab cache for a given combination of allocation flags and size.
206 *
207 * This ought to end up with a global pointer to the right cache
208 * in kmalloc_caches.
209 */
Christoph Lameteraa137f92007-08-31 00:48:45 -0700210static __always_inline struct kmem_cache *kmalloc_slab(size_t size)
Christoph Lameter81819f02007-05-06 14:49:36 -0700211{
212 int index = kmalloc_index(size);
213
214 if (index == 0)
215 return NULL;
216
Christoph Lameter51df1142010-08-20 12:37:15 -0500217 return kmalloc_caches[index];
Christoph Lameter81819f02007-05-06 14:49:36 -0700218}
219
Paul Mundt6193a2f2007-07-15 23:38:22 -0700220void *kmem_cache_alloc(struct kmem_cache *, gfp_t);
221void *__kmalloc(size_t size, gfp_t flags);
222
Richard Kennedy4a923792010-10-21 10:29:19 +0100223static __always_inline void *
224kmalloc_order(size_t size, gfp_t flags, unsigned int order)
225{
226 void *ret = (void *) __get_free_pages(flags | __GFP_COMP, order);
227 kmemleak_alloc(ret, size, 1, flags);
228 return ret;
229}
230
Li Zefan0f24f122009-12-11 15:45:30 +0800231#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +0100232extern void *
233kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size);
234extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300235#else
236static __always_inline void *
Richard Kennedy4a923792010-10-21 10:29:19 +0100237kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300238{
239 return kmem_cache_alloc(s, gfpflags);
240}
Richard Kennedy4a923792010-10-21 10:29:19 +0100241
242static __always_inline void *
243kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
244{
245 return kmalloc_order(size, flags, order);
246}
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300247#endif
248
Pekka Enbergeada35e2008-02-11 22:47:46 +0200249static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
250{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300251 unsigned int order = get_order(size);
Richard Kennedy4a923792010-10-21 10:29:19 +0100252 return kmalloc_order_trace(size, flags, order);
Pekka Enbergeada35e2008-02-11 22:47:46 +0200253}
254
Christoph Lameteraa137f92007-08-31 00:48:45 -0700255static __always_inline void *kmalloc(size_t size, gfp_t flags)
Christoph Lameter81819f02007-05-06 14:49:36 -0700256{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700257 if (__builtin_constant_p(size)) {
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500258 if (size > SLUB_MAX_SIZE)
Pekka Enbergeada35e2008-02-11 22:47:46 +0200259 return kmalloc_large(size, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -0700260
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700261 if (!(flags & SLUB_DMA)) {
262 struct kmem_cache *s = kmalloc_slab(size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700263
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700264 if (!s)
265 return ZERO_SIZE_PTR;
266
Richard Kennedy4a923792010-10-21 10:29:19 +0100267 return kmem_cache_alloc_trace(s, flags, size);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700268 }
269 }
270 return __kmalloc(size, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -0700271}
272
Christoph Lameter81819f02007-05-06 14:49:36 -0700273#ifdef CONFIG_NUMA
Paul Mundt6193a2f2007-07-15 23:38:22 -0700274void *__kmalloc_node(size_t size, gfp_t flags, int node);
275void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node);
Christoph Lameter81819f02007-05-06 14:49:36 -0700276
Li Zefan0f24f122009-12-11 15:45:30 +0800277#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +0100278extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300279 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +0100280 int node, size_t size);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300281#else
282static __always_inline void *
Richard Kennedy4a923792010-10-21 10:29:19 +0100283kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300284 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +0100285 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300286{
287 return kmem_cache_alloc_node(s, gfpflags, node);
288}
289#endif
290
Christoph Lameteraa137f92007-08-31 00:48:45 -0700291static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
Christoph Lameter81819f02007-05-06 14:49:36 -0700292{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700293 if (__builtin_constant_p(size) &&
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500294 size <= SLUB_MAX_SIZE && !(flags & SLUB_DMA)) {
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700295 struct kmem_cache *s = kmalloc_slab(size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700296
297 if (!s)
Christoph Lameter272c1d22007-06-08 13:46:49 -0700298 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -0700299
Richard Kennedy4a923792010-10-21 10:29:19 +0100300 return kmem_cache_alloc_node_trace(s, flags, node, size);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700301 }
302 return __kmalloc_node(size, flags, node);
Christoph Lameter81819f02007-05-06 14:49:36 -0700303}
304#endif
305
Christoph Lameter81819f02007-05-06 14:49:36 -0700306#endif /* _LINUX_SLUB_DEF_H */