blob: 9f63538928c0451c23b02cfe00c309ea00d54534 [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>
Catalin Marinase4f7c0b42009-07-07 10:32:59 +010013#include <linux/kmemleak.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070014
Li Zefan039ca4e2010-05-26 17:22:17 +080015#include <trace/events/kmem.h>
16
Christoph Lameter8ff12cf2008-02-07 17:47:41 -080017enum stat_item {
18 ALLOC_FASTPATH, /* Allocation from cpu slab */
19 ALLOC_SLOWPATH, /* Allocation by getting a new cpu slab */
20 FREE_FASTPATH, /* Free to cpu slub */
21 FREE_SLOWPATH, /* Freeing not to cpu slab */
22 FREE_FROZEN, /* Freeing to frozen slab */
23 FREE_ADD_PARTIAL, /* Freeing moves slab to partial list */
24 FREE_REMOVE_PARTIAL, /* Freeing removes last object */
25 ALLOC_FROM_PARTIAL, /* Cpu slab acquired from partial list */
26 ALLOC_SLAB, /* Cpu slab acquired from page allocator */
27 ALLOC_REFILL, /* Refill cpu slab from slab freelist */
28 FREE_SLAB, /* Slab freed to the page allocator */
29 CPUSLAB_FLUSH, /* Abandoning of the cpu slab */
30 DEACTIVATE_FULL, /* Cpu slab was full when deactivated */
31 DEACTIVATE_EMPTY, /* Cpu slab was empty when deactivated */
32 DEACTIVATE_TO_HEAD, /* Cpu slab was moved to the head of partials */
33 DEACTIVATE_TO_TAIL, /* Cpu slab was moved to the tail of partials */
34 DEACTIVATE_REMOTE_FREES,/* Slab contained remotely freed objects */
Christoph Lameter65c33762008-04-14 19:11:40 +030035 ORDER_FALLBACK, /* Number of times fallback was necessary */
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 Lameterda89b792008-01-07 23:20:31 -080039 void **freelist; /* Pointer to first free per cpu object */
40 struct page *page; /* The slab from which we are allocating */
41 int node; /* The node of the page (or -1 for debug) */
Christoph Lameter8ff12cf2008-02-07 17:47:41 -080042#ifdef CONFIG_SLUB_STATS
43 unsigned stat[NR_SLUB_STAT_ITEMS];
44#endif
Christoph Lameter4c93c3552007-10-16 01:26:08 -070045};
Christoph Lameterdfb4f092007-10-16 01:26:05 -070046
Christoph Lameter81819f02007-05-06 14:49:36 -070047struct kmem_cache_node {
48 spinlock_t list_lock; /* Protect partial list and nr_partial */
49 unsigned long nr_partial;
Christoph Lameter81819f02007-05-06 14:49:36 -070050 struct list_head partial;
Christoph Lameter0c710012007-07-17 04:03:24 -070051#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +030052 atomic_long_t nr_slabs;
Christoph Lameter205ab992008-04-14 19:11:40 +030053 atomic_long_t total_objects;
Christoph Lameter643b1132007-05-06 14:49:42 -070054 struct list_head full;
Christoph Lameter0c710012007-07-17 04:03:24 -070055#endif
Christoph Lameter81819f02007-05-06 14:49:36 -070056};
57
58/*
Christoph Lameter834f3d12008-04-14 19:11:31 +030059 * Word size structure that can be atomically updated or read and that
60 * contains both the order and the number of objects that a slab of the
61 * given order would contain.
62 */
63struct kmem_cache_order_objects {
64 unsigned long x;
65};
66
67/*
Christoph Lameter81819f02007-05-06 14:49:36 -070068 * Slab cache management.
69 */
70struct kmem_cache {
Namhyung Kim1b5ad242010-08-07 14:29:22 +020071 struct kmem_cache_cpu __percpu *cpu_slab;
Christoph Lameter81819f02007-05-06 14:49:36 -070072 /* Used for retriving partial slabs etc */
73 unsigned long flags;
74 int size; /* The size of an object including meta data */
75 int objsize; /* The size of an object without meta data */
76 int offset; /* Free pointer offset. */
Christoph Lameter834f3d12008-04-14 19:11:31 +030077 struct kmem_cache_order_objects oo;
Christoph Lameter81819f02007-05-06 14:49:36 -070078
Christoph Lameter81819f02007-05-06 14:49:36 -070079 /* Allocation and freeing of slabs */
Christoph Lameter205ab992008-04-14 19:11:40 +030080 struct kmem_cache_order_objects max;
Christoph Lameter65c33762008-04-14 19:11:40 +030081 struct kmem_cache_order_objects min;
Christoph Lameterb7a49f02008-02-14 14:21:32 -080082 gfp_t allocflags; /* gfp flags to use on each alloc */
Christoph Lameter81819f02007-05-06 14:49:36 -070083 int refcount; /* Refcount for slab cache destroy */
Alexey Dobriyan51cc5062008-07-25 19:45:34 -070084 void (*ctor)(void *);
Christoph Lameter81819f02007-05-06 14:49:36 -070085 int inuse; /* Offset to metadata */
86 int align; /* Alignment */
David Rientjes3b89d7d2009-02-22 17:40:07 -080087 unsigned long min_partial;
Christoph Lameter81819f02007-05-06 14:49:36 -070088 const char *name; /* Name (only for display!) */
89 struct list_head list; /* List of slab caches */
Christoph Lameter0c710012007-07-17 04:03:24 -070090#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -070091 struct kobject kobj; /* For sysfs */
Christoph Lameter0c710012007-07-17 04:03:24 -070092#endif
Christoph Lameter81819f02007-05-06 14:49:36 -070093
94#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -080095 /*
96 * Defragmentation by allocating from a remote node.
97 */
98 int remote_node_defrag_ratio;
Christoph Lameter81819f02007-05-06 14:49:36 -070099 struct kmem_cache_node *node[MAX_NUMNODES];
Alexander Duyck73367bd2010-05-21 14:41:35 -0700100#else
101 /* Avoid an extra cache line for UP */
102 struct kmem_cache_node local_node;
Christoph Lameter81819f02007-05-06 14:49:36 -0700103#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700104};
105
106/*
107 * Kmalloc subsystem.
108 */
FUJITA Tomonoria6eb9fe2010-08-10 18:03:22 -0700109#if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
110#define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
Christoph Lameter4b356be2007-06-16 10:16:13 -0700111#else
112#define KMALLOC_MIN_SIZE 8
113#endif
114
115#define KMALLOC_SHIFT_LOW ilog2(KMALLOC_MIN_SIZE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700116
FUJITA Tomonoria6eb9fe2010-08-10 18:03:22 -0700117#ifdef ARCH_DMA_MINALIGN
118#define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
119#else
David Woodhouse4581ced2010-05-19 12:02:14 +0100120#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
121#endif
122
123#ifndef ARCH_SLAB_MINALIGN
124#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
125#endif
126
Christoph Lameter81819f02007-05-06 14:49:36 -0700127/*
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500128 * Maximum kmalloc object size handled by SLUB. Larger object allocations
129 * are passed through to the page allocator. The page allocator "fastpath"
130 * is relatively slow so we need this value sufficiently high so that
131 * performance critical objects are allocated through the SLUB fastpath.
132 *
133 * This should be dropped to PAGE_SIZE / 2 once the page allocator
134 * "fastpath" becomes competitive with the slab allocator fastpaths.
135 */
Pekka Enberg51735a72009-02-20 12:21:33 +0200136#define SLUB_MAX_SIZE (2 * PAGE_SIZE)
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500137
Pekka Enberg51735a72009-02-20 12:21:33 +0200138#define SLUB_PAGE_SHIFT (PAGE_SHIFT + 2)
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500139
Christoph Lameter756dee72009-12-18 16:26:21 -0600140#ifdef CONFIG_ZONE_DMA
141#define SLUB_DMA __GFP_DMA
142/* Reserve extra caches for potential DMA use */
Christoph Lameter0f1f6942010-05-27 09:17:17 -0500143#define KMALLOC_CACHES (2 * SLUB_PAGE_SHIFT)
Christoph Lameter756dee72009-12-18 16:26:21 -0600144#else
145/* Disable DMA functionality */
146#define SLUB_DMA (__force gfp_t)0
147#define KMALLOC_CACHES SLUB_PAGE_SHIFT
148#endif
149
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500150/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700151 * We keep the general caches in an array of slab caches that are used for
152 * 2^x bytes of allocations.
153 */
Christoph Lameter756dee72009-12-18 16:26:21 -0600154extern struct kmem_cache kmalloc_caches[KMALLOC_CACHES];
Christoph Lameter81819f02007-05-06 14:49:36 -0700155
156/*
157 * Sorry that the following has to be that ugly but some versions of GCC
158 * have trouble with constant propagation and loops.
159 */
Christoph Lameteraa137f92007-08-31 00:48:45 -0700160static __always_inline int kmalloc_index(size_t size)
Christoph Lameter81819f02007-05-06 14:49:36 -0700161{
Christoph Lameter272c1d22007-06-08 13:46:49 -0700162 if (!size)
163 return 0;
Christoph Lameter614410d2007-05-06 14:49:38 -0700164
Christoph Lameter4b356be2007-06-16 10:16:13 -0700165 if (size <= KMALLOC_MIN_SIZE)
166 return KMALLOC_SHIFT_LOW;
167
Aaro Koskinenacdfcd02009-08-28 14:28:54 +0300168 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
Christoph Lameter81819f02007-05-06 14:49:36 -0700169 return 1;
Aaro Koskinenacdfcd02009-08-28 14:28:54 +0300170 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
Christoph Lameter81819f02007-05-06 14:49:36 -0700171 return 2;
172 if (size <= 8) return 3;
173 if (size <= 16) return 4;
174 if (size <= 32) return 5;
175 if (size <= 64) return 6;
176 if (size <= 128) return 7;
177 if (size <= 256) return 8;
178 if (size <= 512) return 9;
179 if (size <= 1024) return 10;
180 if (size <= 2 * 1024) return 11;
Christoph Lameter6446faa2008-02-15 23:45:26 -0800181 if (size <= 4 * 1024) return 12;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700182/*
183 * The following is only needed to support architectures with a larger page
184 * size than 4k.
185 */
Christoph Lameter81819f02007-05-06 14:49:36 -0700186 if (size <= 8 * 1024) return 13;
187 if (size <= 16 * 1024) return 14;
188 if (size <= 32 * 1024) return 15;
189 if (size <= 64 * 1024) return 16;
190 if (size <= 128 * 1024) return 17;
191 if (size <= 256 * 1024) return 18;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700192 if (size <= 512 * 1024) return 19;
Christoph Lameter81819f02007-05-06 14:49:36 -0700193 if (size <= 1024 * 1024) return 20;
Christoph Lameter81819f02007-05-06 14:49:36 -0700194 if (size <= 2 * 1024 * 1024) return 21;
Christoph Lameter81819f02007-05-06 14:49:36 -0700195 return -1;
196
197/*
198 * What we really wanted to do and cannot do because of compiler issues is:
199 * int i;
200 * for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++)
201 * if (size <= (1 << i))
202 * return i;
203 */
204}
205
206/*
207 * Find the slab cache for a given combination of allocation flags and size.
208 *
209 * This ought to end up with a global pointer to the right cache
210 * in kmalloc_caches.
211 */
Christoph Lameteraa137f92007-08-31 00:48:45 -0700212static __always_inline struct kmem_cache *kmalloc_slab(size_t size)
Christoph Lameter81819f02007-05-06 14:49:36 -0700213{
214 int index = kmalloc_index(size);
215
216 if (index == 0)
217 return NULL;
218
Christoph Lameter81819f02007-05-06 14:49:36 -0700219 return &kmalloc_caches[index];
220}
221
Paul Mundt6193a2f2007-07-15 23:38:22 -0700222void *kmem_cache_alloc(struct kmem_cache *, gfp_t);
223void *__kmalloc(size_t size, gfp_t flags);
224
Li Zefan0f24f122009-12-11 15:45:30 +0800225#ifdef CONFIG_TRACING
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300226extern void *kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags);
227#else
228static __always_inline void *
229kmem_cache_alloc_notrace(struct kmem_cache *s, gfp_t gfpflags)
230{
231 return kmem_cache_alloc(s, gfpflags);
232}
233#endif
234
Pekka Enbergeada35e2008-02-11 22:47:46 +0200235static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
236{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300237 unsigned int order = get_order(size);
238 void *ret = (void *) __get_free_pages(flags | __GFP_COMP, order);
239
Catalin Marinase4f7c0b42009-07-07 10:32:59 +0100240 kmemleak_alloc(ret, size, 1, flags);
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +0200241 trace_kmalloc(_THIS_IP_, ret, size, PAGE_SIZE << order, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300242
243 return ret;
Pekka Enbergeada35e2008-02-11 22:47:46 +0200244}
245
Christoph Lameteraa137f92007-08-31 00:48:45 -0700246static __always_inline void *kmalloc(size_t size, gfp_t flags)
Christoph Lameter81819f02007-05-06 14:49:36 -0700247{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300248 void *ret;
249
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700250 if (__builtin_constant_p(size)) {
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500251 if (size > SLUB_MAX_SIZE)
Pekka Enbergeada35e2008-02-11 22:47:46 +0200252 return kmalloc_large(size, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -0700253
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700254 if (!(flags & SLUB_DMA)) {
255 struct kmem_cache *s = kmalloc_slab(size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700256
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700257 if (!s)
258 return ZERO_SIZE_PTR;
259
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300260 ret = kmem_cache_alloc_notrace(s, flags);
261
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +0200262 trace_kmalloc(_THIS_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300263
264 return ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700265 }
266 }
267 return __kmalloc(size, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -0700268}
269
Christoph Lameter81819f02007-05-06 14:49:36 -0700270#ifdef CONFIG_NUMA
Paul Mundt6193a2f2007-07-15 23:38:22 -0700271void *__kmalloc_node(size_t size, gfp_t flags, int node);
272void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node);
Christoph Lameter81819f02007-05-06 14:49:36 -0700273
Li Zefan0f24f122009-12-11 15:45:30 +0800274#ifdef CONFIG_TRACING
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300275extern void *kmem_cache_alloc_node_notrace(struct kmem_cache *s,
276 gfp_t gfpflags,
277 int node);
278#else
279static __always_inline void *
280kmem_cache_alloc_node_notrace(struct kmem_cache *s,
281 gfp_t gfpflags,
282 int node)
283{
284 return kmem_cache_alloc_node(s, gfpflags, node);
285}
286#endif
287
Christoph Lameteraa137f92007-08-31 00:48:45 -0700288static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
Christoph Lameter81819f02007-05-06 14:49:36 -0700289{
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300290 void *ret;
291
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700292 if (__builtin_constant_p(size) &&
Christoph Lameterffadd4d2009-02-17 12:05:07 -0500293 size <= SLUB_MAX_SIZE && !(flags & SLUB_DMA)) {
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700294 struct kmem_cache *s = kmalloc_slab(size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700295
296 if (!s)
Christoph Lameter272c1d22007-06-08 13:46:49 -0700297 return ZERO_SIZE_PTR;
Christoph Lameter81819f02007-05-06 14:49:36 -0700298
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300299 ret = kmem_cache_alloc_node_notrace(s, flags, node);
300
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +0200301 trace_kmalloc_node(_THIS_IP_, ret,
302 size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +0300303
304 return ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -0700305 }
306 return __kmalloc_node(size, flags, node);
Christoph Lameter81819f02007-05-06 14:49:36 -0700307}
308#endif
309
Christoph Lameter81819f02007-05-06 14:49:36 -0700310#endif /* _LINUX_SLUB_DEF_H */