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> | |
187f1882 | 11 | #include <linux/bug.h> |
81819f0f CL |
12 | #include <linux/workqueue.h> |
13 | #include <linux/kobject.h> | |
14 | ||
4a92379b | 15 | #include <linux/kmemleak.h> |
039ca4e7 | 16 | |
8ff12cfc CL |
17 | enum 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 */ | |
8028dcea | 25 | ALLOC_FROM_PARTIAL, /* Cpu slab acquired from node partial list */ |
8ff12cfc CL |
26 | ALLOC_SLAB, /* Cpu slab acquired from page allocator */ |
27 | ALLOC_REFILL, /* Refill cpu slab from slab freelist */ | |
e36a2652 | 28 | ALLOC_NODE_MISMATCH, /* Switching cpu slab */ |
8ff12cfc CL |
29 | FREE_SLAB, /* Slab freed to the page allocator */ |
30 | CPUSLAB_FLUSH, /* Abandoning of the cpu slab */ | |
31 | DEACTIVATE_FULL, /* Cpu slab was full when deactivated */ | |
32 | DEACTIVATE_EMPTY, /* Cpu slab was empty when deactivated */ | |
33 | DEACTIVATE_TO_HEAD, /* Cpu slab was moved to the head of partials */ | |
34 | DEACTIVATE_TO_TAIL, /* Cpu slab was moved to the tail of partials */ | |
35 | DEACTIVATE_REMOTE_FREES,/* Slab contained remotely freed objects */ | |
03e404af | 36 | DEACTIVATE_BYPASS, /* Implicit deactivation */ |
65c3376a | 37 | ORDER_FALLBACK, /* Number of times fallback was necessary */ |
4fdccdfb | 38 | CMPXCHG_DOUBLE_CPU_FAIL,/* Failure of this_cpu_cmpxchg_double */ |
b789ef51 | 39 | CMPXCHG_DOUBLE_FAIL, /* Number of times that cmpxchg double did not match */ |
49e22585 | 40 | CPU_PARTIAL_ALLOC, /* Used cpu partial on alloc */ |
8028dcea AS |
41 | CPU_PARTIAL_FREE, /* Refill cpu partial on free */ |
42 | CPU_PARTIAL_NODE, /* Refill cpu partial from node partial */ | |
43 | CPU_PARTIAL_DRAIN, /* Drain cpu partial to node partial */ | |
8ff12cfc CL |
44 | NR_SLUB_STAT_ITEMS }; |
45 | ||
dfb4f096 | 46 | struct kmem_cache_cpu { |
8a5ec0ba | 47 | void **freelist; /* Pointer to next available object */ |
8a5ec0ba | 48 | unsigned long tid; /* Globally unique transaction id */ |
da89b79e | 49 | struct page *page; /* The slab from which we are allocating */ |
49e22585 | 50 | struct page *partial; /* Partially allocated frozen slabs */ |
8ff12cfc CL |
51 | #ifdef CONFIG_SLUB_STATS |
52 | unsigned stat[NR_SLUB_STAT_ITEMS]; | |
53 | #endif | |
4c93c355 | 54 | }; |
dfb4f096 | 55 | |
81819f0f CL |
56 | struct kmem_cache_node { |
57 | spinlock_t list_lock; /* Protect partial list and nr_partial */ | |
58 | unsigned long nr_partial; | |
81819f0f | 59 | struct list_head partial; |
0c710013 | 60 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 61 | atomic_long_t nr_slabs; |
205ab99d | 62 | atomic_long_t total_objects; |
643b1138 | 63 | struct list_head full; |
0c710013 | 64 | #endif |
81819f0f CL |
65 | }; |
66 | ||
834f3d11 CL |
67 | /* |
68 | * Word size structure that can be atomically updated or read and that | |
69 | * contains both the order and the number of objects that a slab of the | |
70 | * given order would contain. | |
71 | */ | |
72 | struct kmem_cache_order_objects { | |
73 | unsigned long x; | |
74 | }; | |
75 | ||
81819f0f CL |
76 | /* |
77 | * Slab cache management. | |
78 | */ | |
79 | struct kmem_cache { | |
1b5ad248 | 80 | struct kmem_cache_cpu __percpu *cpu_slab; |
81819f0f CL |
81 | /* Used for retriving partial slabs etc */ |
82 | unsigned long flags; | |
1a757fe5 | 83 | unsigned long min_partial; |
81819f0f | 84 | int size; /* The size of an object including meta data */ |
3b0efdfa | 85 | int object_size; /* The size of an object without meta data */ |
81819f0f | 86 | int offset; /* Free pointer offset. */ |
9f264904 | 87 | int cpu_partial; /* Number of per cpu partial objects to keep around */ |
834f3d11 | 88 | struct kmem_cache_order_objects oo; |
81819f0f | 89 | |
81819f0f | 90 | /* Allocation and freeing of slabs */ |
205ab99d | 91 | struct kmem_cache_order_objects max; |
65c3376a | 92 | struct kmem_cache_order_objects min; |
b7a49f0d | 93 | gfp_t allocflags; /* gfp flags to use on each alloc */ |
81819f0f | 94 | int refcount; /* Refcount for slab cache destroy */ |
51cc5068 | 95 | void (*ctor)(void *); |
81819f0f CL |
96 | int inuse; /* Offset to metadata */ |
97 | int align; /* Alignment */ | |
ab9a0f19 | 98 | int reserved; /* Reserved bytes at the end of slabs */ |
81819f0f CL |
99 | const char *name; /* Name (only for display!) */ |
100 | struct list_head list; /* List of slab caches */ | |
ab4d5ed5 | 101 | #ifdef CONFIG_SYSFS |
81819f0f | 102 | struct kobject kobj; /* For sysfs */ |
0c710013 | 103 | #endif |
ba6c496e GC |
104 | #ifdef CONFIG_MEMCG_KMEM |
105 | struct memcg_cache_params *memcg_params; | |
107dab5c | 106 | int max_attr_size; /* for propagation, maximum size of a stored attr */ |
ba6c496e | 107 | #endif |
81819f0f CL |
108 | |
109 | #ifdef CONFIG_NUMA | |
9824601e CL |
110 | /* |
111 | * Defragmentation by allocating from a remote node. | |
112 | */ | |
113 | int remote_node_defrag_ratio; | |
81819f0f | 114 | #endif |
7340cc84 | 115 | struct kmem_cache_node *node[MAX_NUMNODES]; |
81819f0f CL |
116 | }; |
117 | ||
118 | /* | |
119 | * Kmalloc subsystem. | |
120 | */ | |
a6eb9fe1 FT |
121 | #if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8 |
122 | #define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN | |
4b356be0 CL |
123 | #else |
124 | #define KMALLOC_MIN_SIZE 8 | |
125 | #endif | |
126 | ||
127 | #define KMALLOC_SHIFT_LOW ilog2(KMALLOC_MIN_SIZE) | |
81819f0f | 128 | |
ffadd4d0 CL |
129 | /* |
130 | * Maximum kmalloc object size handled by SLUB. Larger object allocations | |
131 | * are passed through to the page allocator. The page allocator "fastpath" | |
132 | * is relatively slow so we need this value sufficiently high so that | |
133 | * performance critical objects are allocated through the SLUB fastpath. | |
134 | * | |
135 | * This should be dropped to PAGE_SIZE / 2 once the page allocator | |
136 | * "fastpath" becomes competitive with the slab allocator fastpaths. | |
137 | */ | |
51735a7c | 138 | #define SLUB_MAX_SIZE (2 * PAGE_SIZE) |
ffadd4d0 | 139 | |
51735a7c | 140 | #define SLUB_PAGE_SHIFT (PAGE_SHIFT + 2) |
ffadd4d0 | 141 | |
756dee75 CL |
142 | #ifdef CONFIG_ZONE_DMA |
143 | #define SLUB_DMA __GFP_DMA | |
756dee75 CL |
144 | #else |
145 | /* Disable DMA functionality */ | |
146 | #define SLUB_DMA (__force gfp_t)0 | |
756dee75 CL |
147 | #endif |
148 | ||
81819f0f CL |
149 | /* |
150 | * We keep the general caches in an array of slab caches that are used for | |
151 | * 2^x bytes of allocations. | |
152 | */ | |
51df1142 | 153 | extern struct kmem_cache *kmalloc_caches[SLUB_PAGE_SHIFT]; |
81819f0f CL |
154 | |
155 | /* | |
156 | * Sorry that the following has to be that ugly but some versions of GCC | |
157 | * have trouble with constant propagation and loops. | |
158 | */ | |
aa137f9d | 159 | static __always_inline int kmalloc_index(size_t size) |
81819f0f | 160 | { |
272c1d21 CL |
161 | if (!size) |
162 | return 0; | |
614410d5 | 163 | |
4b356be0 CL |
164 | if (size <= KMALLOC_MIN_SIZE) |
165 | return KMALLOC_SHIFT_LOW; | |
166 | ||
acdfcd04 | 167 | if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96) |
81819f0f | 168 | return 1; |
acdfcd04 | 169 | if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192) |
81819f0f CL |
170 | return 2; |
171 | if (size <= 8) return 3; | |
172 | if (size <= 16) return 4; | |
173 | if (size <= 32) return 5; | |
174 | if (size <= 64) return 6; | |
175 | if (size <= 128) return 7; | |
176 | if (size <= 256) return 8; | |
177 | if (size <= 512) return 9; | |
178 | if (size <= 1024) return 10; | |
179 | if (size <= 2 * 1024) return 11; | |
6446faa2 | 180 | if (size <= 4 * 1024) return 12; |
aadb4bc4 CL |
181 | /* |
182 | * The following is only needed to support architectures with a larger page | |
3e0c2ab6 CL |
183 | * size than 4k. We need to support 2 * PAGE_SIZE here. So for a 64k page |
184 | * size we would have to go up to 128k. | |
aadb4bc4 | 185 | */ |
81819f0f CL |
186 | 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; | |
aadb4bc4 | 192 | if (size <= 512 * 1024) return 19; |
81819f0f | 193 | if (size <= 1024 * 1024) return 20; |
81819f0f | 194 | if (size <= 2 * 1024 * 1024) return 21; |
3e0c2ab6 CL |
195 | BUG(); |
196 | return -1; /* Will never be reached */ | |
81819f0f CL |
197 | |
198 | /* | |
199 | * What we really wanted to do and cannot do because of compiler issues is: | |
200 | * int i; | |
201 | * for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) | |
202 | * if (size <= (1 << i)) | |
203 | * return i; | |
204 | */ | |
205 | } | |
206 | ||
207 | /* | |
208 | * Find the slab cache for a given combination of allocation flags and size. | |
209 | * | |
210 | * This ought to end up with a global pointer to the right cache | |
211 | * in kmalloc_caches. | |
212 | */ | |
aa137f9d | 213 | static __always_inline struct kmem_cache *kmalloc_slab(size_t size) |
81819f0f CL |
214 | { |
215 | int index = kmalloc_index(size); | |
216 | ||
217 | if (index == 0) | |
218 | return NULL; | |
219 | ||
51df1142 | 220 | return kmalloc_caches[index]; |
81819f0f CL |
221 | } |
222 | ||
6193a2ff PM |
223 | void *kmem_cache_alloc(struct kmem_cache *, gfp_t); |
224 | void *__kmalloc(size_t size, gfp_t flags); | |
225 | ||
4a92379b RK |
226 | static __always_inline void * |
227 | kmalloc_order(size_t size, gfp_t flags, unsigned int order) | |
228 | { | |
d79923fa GC |
229 | void *ret; |
230 | ||
231 | flags |= (__GFP_COMP | __GFP_KMEMCG); | |
232 | ret = (void *) __get_free_pages(flags, order); | |
4a92379b RK |
233 | kmemleak_alloc(ret, size, 1, flags); |
234 | return ret; | |
235 | } | |
236 | ||
d18a90dd BG |
237 | /** |
238 | * Calling this on allocated memory will check that the memory | |
239 | * is expected to be in use, and print warnings if not. | |
240 | */ | |
241 | #ifdef CONFIG_SLUB_DEBUG | |
242 | extern bool verify_mem_not_deleted(const void *x); | |
243 | #else | |
244 | static inline bool verify_mem_not_deleted(const void *x) | |
245 | { | |
246 | return true; | |
247 | } | |
248 | #endif | |
249 | ||
0f24f128 | 250 | #ifdef CONFIG_TRACING |
4a92379b RK |
251 | extern void * |
252 | kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size); | |
253 | extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order); | |
5b882be4 EGM |
254 | #else |
255 | static __always_inline void * | |
4a92379b | 256 | kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
5b882be4 EGM |
257 | { |
258 | return kmem_cache_alloc(s, gfpflags); | |
259 | } | |
4a92379b RK |
260 | |
261 | static __always_inline void * | |
262 | kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
263 | { | |
264 | return kmalloc_order(size, flags, order); | |
265 | } | |
5b882be4 EGM |
266 | #endif |
267 | ||
eada35ef PE |
268 | static __always_inline void *kmalloc_large(size_t size, gfp_t flags) |
269 | { | |
5b882be4 | 270 | unsigned int order = get_order(size); |
4a92379b | 271 | return kmalloc_order_trace(size, flags, order); |
eada35ef PE |
272 | } |
273 | ||
aa137f9d | 274 | static __always_inline void *kmalloc(size_t size, gfp_t flags) |
81819f0f | 275 | { |
aadb4bc4 | 276 | if (__builtin_constant_p(size)) { |
ffadd4d0 | 277 | if (size > SLUB_MAX_SIZE) |
eada35ef | 278 | return kmalloc_large(size, flags); |
81819f0f | 279 | |
aadb4bc4 CL |
280 | if (!(flags & SLUB_DMA)) { |
281 | struct kmem_cache *s = kmalloc_slab(size); | |
282 | ||
283 | if (!s) | |
284 | return ZERO_SIZE_PTR; | |
81819f0f | 285 | |
4a92379b | 286 | return kmem_cache_alloc_trace(s, flags, size); |
aadb4bc4 CL |
287 | } |
288 | } | |
289 | return __kmalloc(size, flags); | |
81819f0f CL |
290 | } |
291 | ||
81819f0f | 292 | #ifdef CONFIG_NUMA |
6193a2ff PM |
293 | void *__kmalloc_node(size_t size, gfp_t flags, int node); |
294 | void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node); | |
81819f0f | 295 | |
0f24f128 | 296 | #ifdef CONFIG_TRACING |
4a92379b | 297 | extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 298 | gfp_t gfpflags, |
4a92379b | 299 | int node, size_t size); |
5b882be4 EGM |
300 | #else |
301 | static __always_inline void * | |
4a92379b | 302 | kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 303 | gfp_t gfpflags, |
4a92379b | 304 | int node, size_t size) |
5b882be4 EGM |
305 | { |
306 | return kmem_cache_alloc_node(s, gfpflags, node); | |
307 | } | |
308 | #endif | |
309 | ||
aa137f9d | 310 | static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node) |
81819f0f | 311 | { |
aadb4bc4 | 312 | if (__builtin_constant_p(size) && |
ffadd4d0 | 313 | size <= SLUB_MAX_SIZE && !(flags & SLUB_DMA)) { |
aadb4bc4 | 314 | struct kmem_cache *s = kmalloc_slab(size); |
81819f0f CL |
315 | |
316 | if (!s) | |
272c1d21 | 317 | return ZERO_SIZE_PTR; |
81819f0f | 318 | |
4a92379b | 319 | return kmem_cache_alloc_node_trace(s, flags, node, size); |
aadb4bc4 CL |
320 | } |
321 | return __kmalloc_node(size, flags, node); | |
81819f0f CL |
322 | } |
323 | #endif | |
324 | ||
325 | #endif /* _LINUX_SLUB_DEF_H */ |