blob: fd4fdc72bc8c310f3838babf7b7bfc687be16518 [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 */
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -060040 unsigned long tid; /* Globally unique transaction id */
Christoph Lameterda89b792008-01-07 23:20:31 -080041 struct page *page; /* The slab from which we are allocating */
42 int node; /* The node of the page (or -1 for debug) */
Christoph Lameter8ff12cf2008-02-07 17:47:41 -080043#ifdef CONFIG_SLUB_STATS
44 unsigned stat[NR_SLUB_STAT_ITEMS];
45#endif
Christoph Lameter4c93c3552007-10-16 01:26:08 -070046};
Christoph Lameterdfb4f092007-10-16 01:26:05 -070047
Christoph Lameter81819f02007-05-06 14:49:36 -070048struct kmem_cache_node {
49 spinlock_t list_lock; /* Protect partial list and nr_partial */
50 unsigned long nr_partial;
Christoph Lameter81819f02007-05-06 14:49:36 -070051 struct list_head partial;
Christoph Lameter0c710012007-07-17 04:03:24 -070052#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +030053 atomic_long_t nr_slabs;
Christoph Lameter205ab992008-04-14 19:11:40 +030054 atomic_long_t total_objects;
Christoph Lameter643b1132007-05-06 14:49:42 -070055 struct list_head full;
Christoph Lameter0c710012007-07-17 04:03:24 -070056#endif
Christoph Lameter81819f02007-05-06 14:49:36 -070057};
58
59/*
Christoph Lameter834f3d12008-04-14 19:11:31 +030060 * Word size structure that can be atomically updated or read and that
61 * contains both the order and the number of objects that a slab of the
62 * given order would contain.
63 */
64struct kmem_cache_order_objects {
65 unsigned long x;
66};
67
68/*
Christoph Lameter81819f02007-05-06 14:49:36 -070069 * Slab cache management.
70 */
71struct kmem_cache {
Namhyung Kim1b5ad242010-08-07 14:29:22 +020072 struct kmem_cache_cpu __percpu *cpu_slab;
Christoph Lameter81819f02007-05-06 14:49:36 -070073 /* Used for retriving partial slabs etc */
74 unsigned long flags;
Christoph Lameter1a757fe2011-02-25 11:38:51 -060075 unsigned long min_partial;
Christoph Lameter81819f02007-05-06 14:49:36 -070076 int size; /* The size of an object including meta data */
77 int objsize; /* The size of an object without meta data */
78 int offset; /* Free pointer offset. */
Christoph Lameter834f3d12008-04-14 19:11:31 +030079 struct kmem_cache_order_objects oo;
Christoph Lameter81819f02007-05-06 14:49:36 -070080
Christoph Lameter81819f02007-05-06 14:49:36 -070081 /* Allocation and freeing of slabs */
Christoph Lameter205ab992008-04-14 19:11:40 +030082 struct kmem_cache_order_objects max;
Christoph Lameter65c33762008-04-14 19:11:40 +030083 struct kmem_cache_order_objects min;
Christoph Lameterb7a49f02008-02-14 14:21:32 -080084 gfp_t allocflags; /* gfp flags to use on each alloc */
Christoph Lameter81819f02007-05-06 14:49:36 -070085 int refcount; /* Refcount for slab cache destroy */
Alexey Dobriyan51cc5062008-07-25 19:45:34 -070086 void (*ctor)(void *);
Christoph Lameter81819f02007-05-06 14:49:36 -070087 int inuse; /* Offset to metadata */
88 int align; /* Alignment */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +080089 int reserved; /* Reserved bytes at the end of slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -070090 const char *name; /* Name (only for display!) */
91 struct list_head list; /* List of slab caches */
Christoph Lameterab4d5ed2010-10-05 13:57:26 -050092#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -070093 struct kobject kobj; /* For sysfs */
Christoph Lameter0c710012007-07-17 04:03:24 -070094#endif
Christoph Lameter81819f02007-05-06 14:49:36 -070095
96#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -080097 /*
98 * Defragmentation by allocating from a remote node.
99 */
100 int remote_node_defrag_ratio;
Christoph Lameter81819f02007-05-06 14:49:36 -0700101#endif
Christoph Lameter7340cc82010-09-28 08:10:26 -0500102 struct kmem_cache_node *node[MAX_NUMNODES];
Christoph Lameter81819f02007-05-06 14:49:36 -0700103};
104
105/*
106 * Kmalloc subsystem.
107 */
FUJITA Tomonoria6eb9fe2010-08-10 18:03:22 -0700108#if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
109#define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
Christoph Lameter4b356be2007-06-16 10:16:13 -0700110#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
Christoph Lameter81819f02007-05-06 14:49:36 -0700116/*
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500117 * Maximum kmalloc object size handled by SLUB. Larger object allocations
118 * are passed through to the page allocator. The page allocator "fastpath"
119 * is relatively slow so we need this value sufficiently high so that
120 * performance critical objects are allocated through the SLUB fastpath.
121 *
122 * This should be dropped to PAGE_SIZE / 2 once the page allocator
123 * "fastpath" becomes competitive with the slab allocator fastpaths.
124 */
Pekka Enberg51735a72009-02-20 12:21:33 +0200125#define SLUB_MAX_SIZE (2 * PAGE_SIZE)
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500126
Pekka Enberg51735a72009-02-20 12:21:33 +0200127#define SLUB_PAGE_SHIFT (PAGE_SHIFT + 2)
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500128
Christoph Lameter756dee72009-12-18 16:26:21 -0600129#ifdef CONFIG_ZONE_DMA
130#define SLUB_DMA __GFP_DMA
Christoph Lameter756dee72009-12-18 16:26:21 -0600131#else
132/* Disable DMA functionality */
133#define SLUB_DMA (__force gfp_t)0
Christoph Lameter756dee72009-12-18 16:26:21 -0600134#endif
135
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500136/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700137 * We keep the general caches in an array of slab caches that are used for
138 * 2^x bytes of allocations.
139 */
Christoph Lameter51df1142010-08-20 12:37:15 -0500140extern struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT];
Christoph Lameter81819f02007-05-06 14:49:36 -0700141
142/*
143 * Sorry that the following has to be that ugly but some versions of GCC
144 * have trouble with constant propagation and loops.
145 */
Christoph Lameteraa137f92007-08-31 00:48:45 -0700146static __always_inline int kmalloc_index(size_t size)
Christoph Lameter81819f02007-05-06 14:49:36 -0700147{
Christoph Lameter272c1d22007-06-08 13:46:49 -0700148 if (!size)
149 return 0;
Christoph Lameter614410d2007-05-06 14:49:38 -0700150
Christoph Lameter4b356be2007-06-16 10:16:13 -0700151 if (size <= KMALLOC_MIN_SIZE)
152 return KMALLOC_SHIFT_LOW;
153
Aaro Koskinenacdfcd02009-08-28 14:28:54 +0300154 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
Christoph Lameter81819f02007-05-06 14:49:36 -0700155 return 1;
Aaro Koskinenacdfcd02009-08-28 14:28:54 +0300156 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
Christoph Lameter81819f02007-05-06 14:49:36 -0700157 return 2;
158 if (size <= 8) return 3;
159 if (size <= 16) return 4;
160 if (size <= 32) return 5;
161 if (size <= 64) return 6;
162 if (size <= 128) return 7;
163 if (size <= 256) return 8;
164 if (size <= 512) return 9;
165 if (size <= 1024) return 10;
166 if (size <= 2 * 1024) return 11;
Christoph Lameter6446faa2008-02-15 23:45:26 -0800167 if (size <= 4 * 1024) return 12;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700168/*
169 * The following is only needed to support architectures with a larger page
Christoph Lameter3e0c2ab2011-05-20 09:42:48 -0500170 * size than 4k. We need to support 2 * PAGE_SIZE here. So for a 64k page
171 * size we would have to go up to 128k.
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700172 */
Christoph Lameter81819f02007-05-06 14:49:36 -0700173 if (size <= 8 * 1024) return 13;
174 if (size <= 16 * 1024) return 14;
175 if (size <= 32 * 1024) return 15;
176 if (size <= 64 * 1024) return 16;
177 if (size <= 128 * 1024) return 17;
178 if (size <= 256 * 1024) return 18;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700179 if (size <= 512 * 1024) return 19;
Christoph Lameter81819f02007-05-06 14:49:36 -0700180 if (size <= 1024 * 1024) return 20;
Christoph Lameter81819f02007-05-06 14:49:36 -0700181 if (size <= 2 * 1024 * 1024) return 21;
Christoph Lameter3e0c2ab2011-05-20 09:42:48 -0500182 BUG();
183 return -1; /* Will never be reached */
Christoph Lameter81819f02007-05-06 14:49:36 -0700184
185/*
186 * What we really wanted to do and cannot do because of compiler issues is:
187 * int i;
188 * for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++)
189 * if (size <= (1 << i))
190 * return i;
191 */
192}
193
194/*
195 * Find the slab cache for a given combination of allocation flags and size.
196 *
197 * This ought to end up with a global pointer to the right cache
198 * in kmalloc_caches.
199 */
Christoph Lameteraa137f92007-08-31 00:48:45 -0700200static __always_inline struct kmem_cache *kmalloc_slab(size_t size)
Christoph Lameter81819f02007-05-06 14:49:36 -0700201{
202 int index = kmalloc_index(size);
203
204 if (index == 0)
205 return NULL;
206
Christoph Lameter51df1142010-08-20 12:37:15 -0500207 return kmalloc_caches[index];
Christoph Lameter81819f02007-05-06 14:49:36 -0700208}
209
Paul Mundt6193a2f2007-07-15 23:38:22 -0700210void *kmem_cache_alloc(struct kmem_cache *, gfp_t);
211void *__kmalloc(size_t size, gfp_t flags);
212
Richard Kennedy4a923792010-10-21 10:29:19 +0100213static __always_inline void *
214kmalloc_order(size_t size, gfp_t flags, unsigned int order)
215{
216 void *ret = (void *) __get_free_pages(flags | __GFP_COMP, order);
217 kmemleak_alloc(ret, size, 1, flags);
218 return ret;
219}
220
Li Zefan0f24f122009-12-11 15:45:30 +0800221#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +0100222extern void *
223kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size);
224extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300225#else
226static __always_inline void *
Richard Kennedy4a923792010-10-21 10:29:19 +0100227kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300228{
229 return kmem_cache_alloc(s, gfpflags);
230}
Richard Kennedy4a923792010-10-21 10:29:19 +0100231
232static __always_inline void *
233kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
234{
235 return kmalloc_order(size, flags, order);
236}
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300237#endif
238
Pekka Enbergeada35e2008-02-11 22:47:46 +0200239static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
240{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300241 unsigned int order = get_order(size);
Richard Kennedy4a923792010-10-21 10:29:19 +0100242 return kmalloc_order_trace(size, flags, order);
Pekka Enbergeada35e2008-02-11 22:47:46 +0200243}
244
Christoph Lameteraa137f92007-08-31 00:48:45 -0700245static __always_inline void *kmalloc(size_t size, gfp_t flags)
Christoph Lameter81819f02007-05-06 14:49:36 -0700246{
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
Richard Kennedy4a923792010-10-21 10:29:19 +0100257 return kmem_cache_alloc_trace(s, flags, size);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700258 }
259 }
260 return __kmalloc(size, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -0700261}
262
Christoph Lameter81819f02007-05-06 14:49:36 -0700263#ifdef CONFIG_NUMA
Paul Mundt6193a2f2007-07-15 23:38:22 -0700264void *__kmalloc_node(size_t size, gfp_t flags, int node);
265void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node);
Christoph Lameter81819f02007-05-06 14:49:36 -0700266
Li Zefan0f24f122009-12-11 15:45:30 +0800267#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +0100268extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300269 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +0100270 int node, size_t size);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300271#else
272static __always_inline void *
Richard Kennedy4a923792010-10-21 10:29:19 +0100273kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300274 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +0100275 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300276{
277 return kmem_cache_alloc_node(s, gfpflags, node);
278}
279#endif
280
Christoph Lameteraa137f92007-08-31 00:48:45 -0700281static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
Christoph Lameter81819f02007-05-06 14:49:36 -0700282{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700283 if (__builtin_constant_p(size) &&
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500284 size <= SLUB_MAX_SIZE && !(flags & SLUB_DMA)) {
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700285 struct kmem_cache *s = kmalloc_slab(size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700286
287 if (!s)
Christoph Lameter272c1d22007-06-08 13:46:49 -0700288 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -0700289
Richard Kennedy4a923792010-10-21 10:29:19 +0100290 return kmem_cache_alloc_node_trace(s, flags, node, size);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700291 }
292 return __kmalloc_node(size, flags, node);
Christoph Lameter81819f02007-05-06 14:49:36 -0700293}
294#endif
295
Christoph Lameter81819f02007-05-06 14:49:36 -0700296#endif /* _LINUX_SLUB_DEF_H */