blob: 4ba59cfc1f7562c0bb8cb900c4f8d444554d25d5 [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>
Zhaolei02af61b2009-04-10 14:26:18 +080013#include <linux/kmemtrace.h>
Catalin Marinase4f7c0b42009-07-07 10:32:59 +010014#include <linux/kmemleak.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070015
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 Lameter8ff12cf2008-02-07 17:47:41 -080035 NR_SLUB_STAT_ITEMS };
36
Christoph Lameterdfb4f092007-10-16 01:26:05 -070037struct kmem_cache_cpu {
Christoph Lameterda89b792008-01-07 23:20:31 -080038 void **freelist; /* Pointer to first free per cpu object */
39 struct page *page; /* The slab from which we are allocating */
40 int node; /* The node of the page (or -1 for debug) */
Christoph Lameter8ff12cf2008-02-07 17:47:41 -080041#ifdef CONFIG_SLUB_STATS
42 unsigned stat[NR_SLUB_STAT_ITEMS];
43#endif
Christoph Lameter4c93c3552007-10-16 01:26:08 -070044};
Christoph Lameterdfb4f092007-10-16 01:26:05 -070045
Christoph Lameter81819f02007-05-06 14:49:36 -070046struct kmem_cache_node {
47 spinlock_t list_lock; /* Protect partial list and nr_partial */
48 unsigned long nr_partial;
Christoph Lameter81819f02007-05-06 14:49:36 -070049 struct list_head partial;
Christoph Lameter0c710012007-07-17 04:03:24 -070050#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +030051 atomic_long_t nr_slabs;
Christoph Lameter205ab992008-04-14 19:11:40 +030052 atomic_long_t total_objects;
Christoph Lameter643b1132007-05-06 14:49:42 -070053 struct list_head full;
Christoph Lameter0c710012007-07-17 04:03:24 -070054#endif
Christoph Lameter81819f02007-05-06 14:49:36 -070055};
56
57/*
Christoph Lameter834f3d12008-04-14 19:11:31 +030058 * Word size structure that can be atomically updated or read and that
59 * contains both the order and the number of objects that a slab of the
60 * given order would contain.
61 */
62struct kmem_cache_order_objects {
63 unsigned long x;
64};
65
66/*
Christoph Lameter81819f02007-05-06 14:49:36 -070067 * Slab cache management.
68 */
69struct kmem_cache {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -060070 struct kmem_cache_cpu *cpu_slab;
Christoph Lameter81819f02007-05-06 14:49:36 -070071 /* Used for retriving partial slabs etc */
72 unsigned long flags;
73 int size; /* The size of an object including meta data */
74 int objsize; /* The size of an object without meta data */
75 int offset; /* Free pointer offset. */
Christoph Lameter834f3d12008-04-14 19:11:31 +030076 struct kmem_cache_order_objects oo;
Christoph Lameter81819f02007-05-06 14:49:36 -070077
Christoph Lameter81819f02007-05-06 14:49:36 -070078 /* Allocation and freeing of slabs */
Christoph Lameter205ab992008-04-14 19:11:40 +030079 struct kmem_cache_order_objects max;
Christoph Lameter65c33762008-04-14 19:11:40 +030080 struct kmem_cache_order_objects min;
Christoph Lameterb7a49f02008-02-14 14:21:32 -080081 gfp_t allocflags; /* gfp flags to use on each alloc */
Christoph Lameter81819f02007-05-06 14:49:36 -070082 int refcount; /* Refcount for slab cache destroy */
Alexey Dobriyan51cc5062008-07-25 19:45:34 -070083 void (*ctor)(void *);
Christoph Lameter81819f02007-05-06 14:49:36 -070084 int inuse; /* Offset to metadata */
85 int align; /* Alignment */
David Rientjes3b89d7d2009-02-22 17:40:07 -080086 unsigned long min_partial;
Christoph Lameter81819f02007-05-06 14:49:36 -070087 const char *name; /* Name (only for display!) */
88 struct list_head list; /* List of slab caches */
Christoph Lameter0c710012007-07-17 04:03:24 -070089#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -070090 struct kobject kobj; /* For sysfs */
Christoph Lameter0c710012007-07-17 04:03:24 -070091#endif
Christoph Lameter81819f02007-05-06 14:49:36 -070092
93#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -080094 /*
95 * Defragmentation by allocating from a remote node.
96 */
97 int remote_node_defrag_ratio;
Christoph Lameter81819f02007-05-06 14:49:36 -070098 struct kmem_cache_node *node[MAX_NUMNODES];
Alexander Duyck73367bd2010-05-21 14:41:35 -070099#else
100 /* Avoid an extra cache line for UP */
101 struct kmem_cache_node local_node;
Christoph Lameter81819f02007-05-06 14:49:36 -0700102#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700103};
104
105/*
106 * Kmalloc subsystem.
107 */
Christoph Lameter4b356be2007-06-16 10:16:13 -0700108#if defined(ARCH_KMALLOC_MINALIGN) && ARCH_KMALLOC_MINALIGN > 8
109#define KMALLOC_MIN_SIZE ARCH_KMALLOC_MINALIGN
110#else
111#define KMALLOC_MIN_SIZE 8
112#endif
113
114#define KMALLOC_SHIFT_LOW ilog2(KMALLOC_MIN_SIZE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700115
David Woodhouse4581ced2010-05-19 12:02:14 +0100116#ifndef ARCH_KMALLOC_MINALIGN
117#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
118#endif
119
120#ifndef ARCH_SLAB_MINALIGN
121#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
122#endif
123
Christoph Lameter81819f02007-05-06 14:49:36 -0700124/*
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500125 * Maximum kmalloc object size handled by SLUB. Larger object allocations
126 * are passed through to the page allocator. The page allocator "fastpath"
127 * is relatively slow so we need this value sufficiently high so that
128 * performance critical objects are allocated through the SLUB fastpath.
129 *
130 * This should be dropped to PAGE_SIZE / 2 once the page allocator
131 * "fastpath" becomes competitive with the slab allocator fastpaths.
132 */
Pekka Enberg51735a72009-02-20 12:21:33 +0200133#define SLUB_MAX_SIZE (2 * PAGE_SIZE)
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500134
Pekka Enberg51735a72009-02-20 12:21:33 +0200135#define SLUB_PAGE_SHIFT (PAGE_SHIFT + 2)
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500136
Christoph Lameter756dee72009-12-18 16:26:21 -0600137#ifdef CONFIG_ZONE_DMA
138#define SLUB_DMA __GFP_DMA
139/* Reserve extra caches for potential DMA use */
Christoph Lameter0f1f6942010-05-27 09:17:17 -0500140#define KMALLOC_CACHES (2 * SLUB_PAGE_SHIFT)
Christoph Lameter756dee72009-12-18 16:26:21 -0600141#else
142/* Disable DMA functionality */
143#define SLUB_DMA (__force gfp_t)0
144#define KMALLOC_CACHES SLUB_PAGE_SHIFT
145#endif
146
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500147/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700148 * We keep the general caches in an array of slab caches that are used for
149 * 2^x bytes of allocations.
150 */
Christoph Lameter756dee72009-12-18 16:26:21 -0600151extern struct kmem_cache kmalloc_caches[KMALLOC_CACHES];
Christoph Lameter81819f02007-05-06 14:49:36 -0700152
153/*
154 * Sorry that the following has to be that ugly but some versions of GCC
155 * have trouble with constant propagation and loops.
156 */
Christoph Lameteraa137f92007-08-31 00:48:45 -0700157static __always_inline int kmalloc_index(size_t size)
Christoph Lameter81819f02007-05-06 14:49:36 -0700158{
Christoph Lameter272c1d22007-06-08 13:46:49 -0700159 if (!size)
160 return 0;
Christoph Lameter614410d2007-05-06 14:49:38 -0700161
Christoph Lameter4b356be2007-06-16 10:16:13 -0700162 if (size <= KMALLOC_MIN_SIZE)
163 return KMALLOC_SHIFT_LOW;
164
Aaro Koskinenacdfcd02009-08-28 14:28:54 +0300165 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
Christoph Lameter81819f02007-05-06 14:49:36 -0700166 return 1;
Aaro Koskinenacdfcd02009-08-28 14:28:54 +0300167 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
Christoph Lameter81819f02007-05-06 14:49:36 -0700168 return 2;
169 if (size <= 8) return 3;
170 if (size <= 16) return 4;
171 if (size <= 32) return 5;
172 if (size <= 64) return 6;
173 if (size <= 128) return 7;
174 if (size <= 256) return 8;
175 if (size <= 512) return 9;
176 if (size <= 1024) return 10;
177 if (size <= 2 * 1024) return 11;
Christoph Lameter6446faa2008-02-15 23:45:26 -0800178 if (size <= 4 * 1024) return 12;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700179/*
180 * The following is only needed to support architectures with a larger page
181 * size than 4k.
182 */
Christoph Lameter81819f02007-05-06 14:49:36 -0700183 if (size <= 8 * 1024) return 13;
184 if (size <= 16 * 1024) return 14;
185 if (size <= 32 * 1024) return 15;
186 if (size <= 64 * 1024) return 16;
187 if (size <= 128 * 1024) return 17;
188 if (size <= 256 * 1024) return 18;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700189 if (size <= 512 * 1024) return 19;
Christoph Lameter81819f02007-05-06 14:49:36 -0700190 if (size <= 1024 * 1024) return 20;
Christoph Lameter81819f02007-05-06 14:49:36 -0700191 if (size <= 2 * 1024 * 1024) return 21;
Christoph Lameter81819f02007-05-06 14:49:36 -0700192 return -1;
193
194/*
195 * What we really wanted to do and cannot do because of compiler issues is:
196 * int i;
197 * for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++)
198 * if (size <= (1 << i))
199 * return i;
200 */
201}
202
203/*
204 * Find the slab cache for a given combination of allocation flags and size.
205 *
206 * This ought to end up with a global pointer to the right cache
207 * in kmalloc_caches.
208 */
Christoph Lameteraa137f92007-08-31 00:48:45 -0700209static __always_inline struct kmem_cache *kmalloc_slab(size_t size)
Christoph Lameter81819f02007-05-06 14:49:36 -0700210{
211 int index = kmalloc_index(size);
212
213 if (index == 0)
214 return NULL;
215
Christoph Lameter81819f02007-05-06 14:49:36 -0700216 return &kmalloc_caches[index];
217}
218
Paul Mundt6193a2f2007-07-15 23:38:22 -0700219void *kmem_cache_alloc(struct kmem_cache *, gfp_t);
220void *__kmalloc(size_t size, gfp_t flags);
221
Li Zefan0f24f122009-12-11 15:45:30 +0800222#ifdef CONFIG_TRACING
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300223extern void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags);
224#else
225static __always_inline void *
226kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags)
227{
228 return kmem_cache_alloc(s, gfpflags);
229}
230#endif
231
Pekka Enbergeada35e2008-02-11 22:47:46 +0200232static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
233{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300234 unsigned int order = get_order(size);
235 void *ret = (void *) __get_free_pages(flags | __GFP_COMP, order);
236
Catalin Marinase4f7c0b42009-07-07 10:32:59 +0100237 kmemleak_alloc(ret, size, 1, flags);
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +0200238 trace_kmalloc(_THIS_IP_, ret, size, PAGE_SIZE << order, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300239
240 return ret;
Pekka Enbergeada35e2008-02-11 22:47:46 +0200241}
242
Christoph Lameteraa137f92007-08-31 00:48:45 -0700243static __always_inline void *kmalloc(size_t size, gfp_t flags)
Christoph Lameter81819f02007-05-06 14:49:36 -0700244{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300245 void *ret;
246
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700247 if (__builtin_constant_p(size)) {
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500248 if (size > SLUB_MAX_SIZE)
Pekka Enbergeada35e2008-02-11 22:47:46 +0200249 return kmalloc_large(size, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -0700250
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700251 if (!(flags & SLUB_DMA)) {
252 struct kmem_cache *s = kmalloc_slab(size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700253
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700254 if (!s)
255 return ZERO_SIZE_PTR;
256
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300257 ret = kmem_cache_alloc_notrace(s, flags);
258
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +0200259 trace_kmalloc(_THIS_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300260
261 return ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700262 }
263 }
264 return __kmalloc(size, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -0700265}
266
Christoph Lameter81819f02007-05-06 14:49:36 -0700267#ifdef CONFIG_NUMA
Paul Mundt6193a2f2007-07-15 23:38:22 -0700268void *__kmalloc_node(size_t size, gfp_t flags, int node);
269void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node);
Christoph Lameter81819f02007-05-06 14:49:36 -0700270
Li Zefan0f24f122009-12-11 15:45:30 +0800271#ifdef CONFIG_TRACING
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300272extern void *kmem_cache_alloc_node_notrace(struct kmem_cache *s,
273 gfp_t gfpflags,
274 int node);
275#else
276static __always_inline void *
277kmem_cache_alloc_node_notrace(struct kmem_cache *s,
278 gfp_t gfpflags,
279 int node)
280{
281 return kmem_cache_alloc_node(s, gfpflags, node);
282}
283#endif
284
Christoph Lameteraa137f92007-08-31 00:48:45 -0700285static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
Christoph Lameter81819f02007-05-06 14:49:36 -0700286{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300287 void *ret;
288
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700289 if (__builtin_constant_p(size) &&
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500290 size <= SLUB_MAX_SIZE && !(flags & SLUB_DMA)) {
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700291 struct kmem_cache *s = kmalloc_slab(size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700292
293 if (!s)
Christoph Lameter272c1d22007-06-08 13:46:49 -0700294 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -0700295
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300296 ret = kmem_cache_alloc_node_notrace(s, flags, node);
297
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +0200298 trace_kmalloc_node(_THIS_IP_, ret,
299 size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300300
301 return ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700302 }
303 return __kmalloc_node(size, flags, node);
Christoph Lameter81819f02007-05-06 14:49:36 -0700304}
305#endif
306
Christoph Lameter81819f02007-05-06 14:49:36 -0700307#endif /* _LINUX_SLUB_DEF_H */