Commit | Line | Data |
---|---|---|
81819f0f CL |
1 | #ifndef _LINUX_SLUB_DEF_H |
2 | #define _LINUX_SLUB_DEF_H | |
3 | ||
4 | /* | |
5 | * SLUB : A Slab allocator without object queues. | |
6 | * | |
cde53535 | 7 | * (C) 2007 SGI, Christoph Lameter |
81819f0f CL |
8 | */ |
9 | #include <linux/types.h> | |
10 | #include <linux/gfp.h> | |
11 | #include <linux/workqueue.h> | |
12 | #include <linux/kobject.h> | |
13 | ||
4a92379b | 14 | #include <linux/kmemleak.h> |
039ca4e7 | 15 | |
8ff12cfc CL |
16 | enum 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 */ | |
e36a2652 | 27 | ALLOC_NODE_MISMATCH, /* Switching cpu slab */ |
8ff12cfc CL |
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 */ | |
03e404af | 35 | DEACTIVATE_BYPASS, /* Implicit deactivation */ |
65c3376a | 36 | ORDER_FALLBACK, /* Number of times fallback was necessary */ |
4fdccdfb | 37 | CMPXCHG_DOUBLE_CPU_FAIL,/* Failure of this_cpu_cmpxchg_double */ |
b789ef51 | 38 | CMPXCHG_DOUBLE_FAIL, /* Number of times that cmpxchg double did not match */ |
49e22585 CL |
39 | CPU_PARTIAL_ALLOC, /* Used cpu partial on alloc */ |
40 | CPU_PARTIAL_FREE, /* USed cpu partial on free */ | |
8ff12cfc CL |
41 | NR_SLUB_STAT_ITEMS }; |
42 | ||
dfb4f096 | 43 | struct kmem_cache_cpu { |
8a5ec0ba | 44 | void **freelist; /* Pointer to next available object */ |
8a5ec0ba | 45 | unsigned long tid; /* Globally unique transaction id */ |
da89b79e | 46 | struct page *page; /* The slab from which we are allocating */ |
49e22585 | 47 | struct page *partial; /* Partially allocated frozen slabs */ |
da89b79e | 48 | int node; /* The node of the page (or -1 for debug) */ |
8ff12cfc CL |
49 | #ifdef CONFIG_SLUB_STATS |
50 | unsigned stat[NR_SLUB_STAT_ITEMS]; | |
51 | #endif | |
4c93c355 | 52 | }; |
dfb4f096 | 53 | |
81819f0f CL |
54 | struct kmem_cache_node { |
55 | spinlock_t list_lock; /* Protect partial list and nr_partial */ | |
56 | unsigned long nr_partial; | |
81819f0f | 57 | struct list_head partial; |
0c710013 | 58 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 59 | atomic_long_t nr_slabs; |
205ab99d | 60 | atomic_long_t total_objects; |
643b1138 | 61 | struct list_head full; |
0c710013 | 62 | #endif |
81819f0f CL |
63 | }; |
64 | ||
834f3d11 CL |
65 | /* |
66 | * Word size structure that can be atomically updated or read and that | |
67 | * contains both the order and the number of objects that a slab of the | |
68 | * given order would contain. | |
69 | */ | |
70 | struct kmem_cache_order_objects { | |
71 | unsigned long x; | |
72 | }; | |
73 | ||
81819f0f CL |
74 | /* |
75 | * Slab cache management. | |
76 | */ | |
77 | struct kmem_cache { | |
1b5ad248 | 78 | struct kmem_cache_cpu __percpu *cpu_slab; |
81819f0f CL |
79 | /* Used for retriving partial slabs etc */ |
80 | unsigned long flags; | |
1a757fe5 | 81 | unsigned long min_partial; |
81819f0f CL |
82 | int size; /* The size of an object including meta data */ |
83 | int objsize; /* The size of an object without meta data */ | |
84 | int offset; /* Free pointer offset. */ | |
9f264904 | 85 | int cpu_partial; /* Number of per cpu partial objects to keep around */ |
834f3d11 | 86 | struct kmem_cache_order_objects oo; |
81819f0f | 87 | |
81819f0f | 88 | /* Allocation and freeing of slabs */ |
205ab99d | 89 | struct kmem_cache_order_objects max; |
65c3376a | 90 | struct kmem_cache_order_objects min; |
b7a49f0d | 91 | gfp_t allocflags; /* gfp flags to use on each alloc */ |
81819f0f | 92 | int refcount; /* Refcount for slab cache destroy */ |
51cc5068 | 93 | void (*ctor)(void *); |
81819f0f CL |
94 | int inuse; /* Offset to metadata */ |
95 | int align; /* Alignment */ | |
ab9a0f19 | 96 | int reserved; /* Reserved bytes at the end of slabs */ |
81819f0f CL |
97 | const char *name; /* Name (only for display!) */ |
98 | struct list_head list; /* List of slab caches */ | |
ab4d5ed5 | 99 | #ifdef CONFIG_SYSFS |
81819f0f | 100 | struct kobject kobj; /* For sysfs */ |
0c710013 | 101 | #endif |
81819f0f CL |
102 | |
103 | #ifdef CONFIG_NUMA | |
9824601e CL |
104 | /* |
105 | * Defragmentation by allocating from a remote node. | |
106 | */ | |
107 | int remote_node_defrag_ratio; | |
81819f0f | 108 | #endif |
7340cc84 | 109 | struct kmem_cache_node *node[MAX_NUMNODES]; |
81819f0f CL |
110 | }; |
111 | ||
112 | /* | |
113 | * Kmalloc subsystem. | |
114 | */ | |
a6eb9fe1 FT |
115 | #if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8 |
116 | #define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN | |
4b356be0 CL |
117 | #else |
118 | #define KMALLOC_MIN_SIZE 8 | |
119 | #endif | |
120 | ||
121 | #define KMALLOC_SHIFT_LOW ilog2(KMALLOC_MIN_SIZE) | |
81819f0f | 122 | |
ffadd4d0 CL |
123 | /* |
124 | * Maximum kmalloc object size handled by SLUB. Larger object allocations | |
125 | * are passed through to the page allocator. The page allocator "fastpath" | |
126 | * is relatively slow so we need this value sufficiently high so that | |
127 | * performance critical objects are allocated through the SLUB fastpath. | |
128 | * | |
129 | * This should be dropped to PAGE_SIZE / 2 once the page allocator | |
130 | * "fastpath" becomes competitive with the slab allocator fastpaths. | |
131 | */ | |
51735a7c | 132 | #define SLUB_MAX_SIZE (2 * PAGE_SIZE) |
ffadd4d0 | 133 | |
51735a7c | 134 | #define SLUB_PAGE_SHIFT (PAGE_SHIFT + 2) |
ffadd4d0 | 135 | |
756dee75 CL |
136 | #ifdef CONFIG_ZONE_DMA |
137 | #define SLUB_DMA __GFP_DMA | |
756dee75 CL |
138 | #else |
139 | /* Disable DMA functionality */ | |
140 | #define SLUB_DMA (__force gfp_t)0 | |
756dee75 CL |
141 | #endif |
142 | ||
81819f0f CL |
143 | /* |
144 | * We keep the general caches in an array of slab caches that are used for | |
145 | * 2^x bytes of allocations. | |
146 | */ | |
51df1142 | 147 | extern struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT]; |
81819f0f CL |
148 | |
149 | /* | |
150 | * Sorry that the following has to be that ugly but some versions of GCC | |
151 | * have trouble with constant propagation and loops. | |
152 | */ | |
aa137f9d | 153 | static __always_inline int kmalloc_index(size_t size) |
81819f0f | 154 | { |
272c1d21 CL |
155 | if (!size) |
156 | return 0; | |
614410d5 | 157 | |
4b356be0 CL |
158 | if (size <= KMALLOC_MIN_SIZE) |
159 | return KMALLOC_SHIFT_LOW; | |
160 | ||
acdfcd04 | 161 | if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96) |
81819f0f | 162 | return 1; |
acdfcd04 | 163 | if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192) |
81819f0f CL |
164 | return 2; |
165 | if (size <= 8) return 3; | |
166 | if (size <= 16) return 4; | |
167 | if (size <= 32) return 5; | |
168 | if (size <= 64) return 6; | |
169 | if (size <= 128) return 7; | |
170 | if (size <= 256) return 8; | |
171 | if (size <= 512) return 9; | |
172 | if (size <= 1024) return 10; | |
173 | if (size <= 2 * 1024) return 11; | |
6446faa2 | 174 | if (size <= 4 * 1024) return 12; |
aadb4bc4 CL |
175 | /* |
176 | * The following is only needed to support architectures with a larger page | |
3e0c2ab6 CL |
177 | * size than 4k. We need to support 2 * PAGE_SIZE here. So for a 64k page |
178 | * size we would have to go up to 128k. | |
aadb4bc4 | 179 | */ |
81819f0f CL |
180 | if (size <= 8 * 1024) return 13; |
181 | if (size <= 16 * 1024) return 14; | |
182 | if (size <= 32 * 1024) return 15; | |
183 | if (size <= 64 * 1024) return 16; | |
184 | if (size <= 128 * 1024) return 17; | |
185 | if (size <= 256 * 1024) return 18; | |
aadb4bc4 | 186 | if (size <= 512 * 1024) return 19; |
81819f0f | 187 | if (size <= 1024 * 1024) return 20; |
81819f0f | 188 | if (size <= 2 * 1024 * 1024) return 21; |
3e0c2ab6 CL |
189 | BUG(); |
190 | return -1; /* Will never be reached */ | |
81819f0f CL |
191 | |
192 | /* | |
193 | * What we really wanted to do and cannot do because of compiler issues is: | |
194 | * int i; | |
195 | * for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) | |
196 | * if (size <= (1 << i)) | |
197 | * return i; | |
198 | */ | |
199 | } | |
200 | ||
201 | /* | |
202 | * Find the slab cache for a given combination of allocation flags and size. | |
203 | * | |
204 | * This ought to end up with a global pointer to the right cache | |
205 | * in kmalloc_caches. | |
206 | */ | |
aa137f9d | 207 | static __always_inline struct kmem_cache *kmalloc_slab(size_t size) |
81819f0f CL |
208 | { |
209 | int index = kmalloc_index(size); | |
210 | ||
211 | if (index == 0) | |
212 | return NULL; | |
213 | ||
51df1142 | 214 | return kmalloc_caches[index]; |
81819f0f CL |
215 | } |
216 | ||
6193a2ff PM |
217 | void *kmem_cache_alloc(struct kmem_cache *, gfp_t); |
218 | void *__kmalloc(size_t size, gfp_t flags); | |
219 | ||
4a92379b RK |
220 | static __always_inline void * |
221 | kmalloc_order(size_t size, gfp_t flags, unsigned int order) | |
222 | { | |
223 | void *ret = (void *) __get_free_pages(flags | __GFP_COMP, order); | |
224 | kmemleak_alloc(ret, size, 1, flags); | |
225 | return ret; | |
226 | } | |
227 | ||
d18a90dd BG |
228 | /** |
229 | * Calling this on allocated memory will check that the memory | |
230 | * is expected to be in use, and print warnings if not. | |
231 | */ | |
232 | #ifdef CONFIG_SLUB_DEBUG | |
233 | extern bool verify_mem_not_deleted(const void *x); | |
234 | #else | |
235 | static inline bool verify_mem_not_deleted(const void *x) | |
236 | { | |
237 | return true; | |
238 | } | |
239 | #endif | |
240 | ||
0f24f128 | 241 | #ifdef CONFIG_TRACING |
4a92379b RK |
242 | extern void * |
243 | kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size); | |
244 | extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order); | |
5b882be4 EGM |
245 | #else |
246 | static __always_inline void * | |
4a92379b | 247 | kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
5b882be4 EGM |
248 | { |
249 | return kmem_cache_alloc(s, gfpflags); | |
250 | } | |
4a92379b RK |
251 | |
252 | static __always_inline void * | |
253 | kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
254 | { | |
255 | return kmalloc_order(size, flags, order); | |
256 | } | |
5b882be4 EGM |
257 | #endif |
258 | ||
eada35ef PE |
259 | static __always_inline void *kmalloc_large(size_t size, gfp_t flags) |
260 | { | |
5b882be4 | 261 | unsigned int order = get_order(size); |
4a92379b | 262 | return kmalloc_order_trace(size, flags, order); |
eada35ef PE |
263 | } |
264 | ||
aa137f9d | 265 | static __always_inline void *kmalloc(size_t size, gfp_t flags) |
81819f0f | 266 | { |
aadb4bc4 | 267 | if (__builtin_constant_p(size)) { |
ffadd4d0 | 268 | if (size > SLUB_MAX_SIZE) |
eada35ef | 269 | return kmalloc_large(size, flags); |
81819f0f | 270 | |
aadb4bc4 CL |
271 | if (!(flags & SLUB_DMA)) { |
272 | struct kmem_cache *s = kmalloc_slab(size); | |
273 | ||
274 | if (!s) | |
275 | return ZERO_SIZE_PTR; | |
81819f0f | 276 | |
4a92379b | 277 | return kmem_cache_alloc_trace(s, flags, size); |
aadb4bc4 CL |
278 | } |
279 | } | |
280 | return __kmalloc(size, flags); | |
81819f0f CL |
281 | } |
282 | ||
81819f0f | 283 | #ifdef CONFIG_NUMA |
6193a2ff PM |
284 | void *__kmalloc_node(size_t size, gfp_t flags, int node); |
285 | void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node); | |
81819f0f | 286 | |
0f24f128 | 287 | #ifdef CONFIG_TRACING |
4a92379b | 288 | extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 289 | gfp_t gfpflags, |
4a92379b | 290 | int node, size_t size); |
5b882be4 EGM |
291 | #else |
292 | static __always_inline void * | |
4a92379b | 293 | kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 294 | gfp_t gfpflags, |
4a92379b | 295 | int node, size_t size) |
5b882be4 EGM |
296 | { |
297 | return kmem_cache_alloc_node(s, gfpflags, node); | |
298 | } | |
299 | #endif | |
300 | ||
aa137f9d | 301 | static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node) |
81819f0f | 302 | { |
aadb4bc4 | 303 | if (__builtin_constant_p(size) && |
ffadd4d0 | 304 | size <= SLUB_MAX_SIZE && !(flags & SLUB_DMA)) { |
aadb4bc4 | 305 | struct kmem_cache *s = kmalloc_slab(size); |
81819f0f CL |
306 | |
307 | if (!s) | |
272c1d21 | 308 | return ZERO_SIZE_PTR; |
81819f0f | 309 | |
4a92379b | 310 | return kmem_cache_alloc_node_trace(s, flags, node, size); |
aadb4bc4 CL |
311 | } |
312 | return __kmalloc_node(size, flags, node); | |
81819f0f CL |
313 | } |
314 | #endif | |
315 | ||
316 | #endif /* _LINUX_SLUB_DEF_H */ |