]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
2e892f43 CL |
2 | * Written by Mark Hemment, 1996 (markhe@nextd.demon.co.uk). |
3 | * | |
cde53535 | 4 | * (C) SGI 2006, Christoph Lameter |
2e892f43 CL |
5 | * Cleaned up and restructured to ease the addition of alternative |
6 | * implementations of SLAB allocators. | |
1da177e4 LT |
7 | */ |
8 | ||
9 | #ifndef _LINUX_SLAB_H | |
10 | #define _LINUX_SLAB_H | |
11 | ||
1b1cec4b | 12 | #include <linux/gfp.h> |
1b1cec4b | 13 | #include <linux/types.h> |
1f458cbf GC |
14 | #include <linux/workqueue.h> |
15 | ||
1da177e4 | 16 | |
2e892f43 CL |
17 | /* |
18 | * Flags to pass to kmem_cache_create(). | |
19 | * The ones marked DEBUG are only valid if CONFIG_SLAB_DEBUG is set. | |
1da177e4 | 20 | */ |
55935a34 | 21 | #define SLAB_DEBUG_FREE 0x00000100UL /* DEBUG: Perform (expensive) checks on free */ |
55935a34 CL |
22 | #define SLAB_RED_ZONE 0x00000400UL /* DEBUG: Red zone objs in a cache */ |
23 | #define SLAB_POISON 0x00000800UL /* DEBUG: Poison objects */ | |
24 | #define SLAB_HWCACHE_ALIGN 0x00002000UL /* Align objs on cache lines */ | |
2e892f43 | 25 | #define SLAB_CACHE_DMA 0x00004000UL /* Use GFP_DMA memory */ |
2e892f43 | 26 | #define SLAB_STORE_USER 0x00010000UL /* DEBUG: Store the last owner for bug hunting */ |
2e892f43 | 27 | #define SLAB_PANIC 0x00040000UL /* Panic if kmem_cache_create() fails */ |
d7de4c1d PZ |
28 | /* |
29 | * SLAB_DESTROY_BY_RCU - **WARNING** READ THIS! | |
30 | * | |
31 | * This delays freeing the SLAB page by a grace period, it does _NOT_ | |
32 | * delay object freeing. This means that if you do kmem_cache_free() | |
33 | * that memory location is free to be reused at any time. Thus it may | |
34 | * be possible to see another object there in the same RCU grace period. | |
35 | * | |
36 | * This feature only ensures the memory location backing the object | |
37 | * stays valid, the trick to using this is relying on an independent | |
38 | * object validation pass. Something like: | |
39 | * | |
40 | * rcu_read_lock() | |
41 | * again: | |
42 | * obj = lockless_lookup(key); | |
43 | * if (obj) { | |
44 | * if (!try_get_ref(obj)) // might fail for free objects | |
45 | * goto again; | |
46 | * | |
47 | * if (obj->key != key) { // not the object we expected | |
48 | * put_ref(obj); | |
49 | * goto again; | |
50 | * } | |
51 | * } | |
52 | * rcu_read_unlock(); | |
53 | * | |
54 | * See also the comment on struct slab_rcu in mm/slab.c. | |
55 | */ | |
2e892f43 | 56 | #define SLAB_DESTROY_BY_RCU 0x00080000UL /* Defer freeing slabs to RCU */ |
101a5001 | 57 | #define SLAB_MEM_SPREAD 0x00100000UL /* Spread some memory over cpuset */ |
81819f0f | 58 | #define SLAB_TRACE 0x00200000UL /* Trace allocations and frees */ |
1da177e4 | 59 | |
30327acf TG |
60 | /* Flag to prevent checks on free */ |
61 | #ifdef CONFIG_DEBUG_OBJECTS | |
62 | # define SLAB_DEBUG_OBJECTS 0x00400000UL | |
63 | #else | |
64 | # define SLAB_DEBUG_OBJECTS 0x00000000UL | |
65 | #endif | |
66 | ||
d5cff635 CM |
67 | #define SLAB_NOLEAKTRACE 0x00800000UL /* Avoid kmemleak tracing */ |
68 | ||
2dff4405 VN |
69 | /* Don't track use of uninitialized memory */ |
70 | #ifdef CONFIG_KMEMCHECK | |
71 | # define SLAB_NOTRACK 0x01000000UL | |
72 | #else | |
73 | # define SLAB_NOTRACK 0x00000000UL | |
74 | #endif | |
4c13dd3b DM |
75 | #ifdef CONFIG_FAILSLAB |
76 | # define SLAB_FAILSLAB 0x02000000UL /* Fault injection mark */ | |
77 | #else | |
78 | # define SLAB_FAILSLAB 0x00000000UL | |
79 | #endif | |
2dff4405 | 80 | |
e12ba74d MG |
81 | /* The following flags affect the page allocator grouping pages by mobility */ |
82 | #define SLAB_RECLAIM_ACCOUNT 0x00020000UL /* Objects are reclaimable */ | |
83 | #define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */ | |
6cb8f913 CL |
84 | /* |
85 | * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests. | |
86 | * | |
87 | * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault. | |
88 | * | |
89 | * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can. | |
90 | * Both make kfree a no-op. | |
91 | */ | |
92 | #define ZERO_SIZE_PTR ((void *)16) | |
93 | ||
1d4ec7b1 | 94 | #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \ |
6cb8f913 CL |
95 | (unsigned long)ZERO_SIZE_PTR) |
96 | ||
3b0efdfa CL |
97 | /* |
98 | * Common fields provided in kmem_cache by all slab allocators | |
99 | * This struct is either used directly by the allocator (SLOB) | |
100 | * or the allocator must include definitions for all fields | |
101 | * provided in kmem_cache_common in their definition of kmem_cache. | |
102 | * | |
103 | * Once we can do anonymous structs (C11 standard) we could put a | |
104 | * anonymous struct definition in these allocators so that the | |
105 | * separate allocations in the kmem_cache structure of SLAB and | |
106 | * SLUB is no longer needed. | |
107 | */ | |
108 | #ifdef CONFIG_SLOB | |
109 | struct kmem_cache { | |
110 | unsigned int object_size;/* The original size of the object */ | |
111 | unsigned int size; /* The aligned/padded/added on size */ | |
112 | unsigned int align; /* Alignment as calculated */ | |
113 | unsigned long flags; /* Active flags on the slab */ | |
114 | const char *name; /* Slab name for sysfs */ | |
115 | int refcount; /* Use counter */ | |
116 | void (*ctor)(void *); /* Called on object slot creation */ | |
117 | struct list_head list; /* List of all slab caches on the system */ | |
118 | }; | |
119 | #endif | |
120 | ||
2633d7a0 | 121 | struct mem_cgroup; |
2e892f43 CL |
122 | /* |
123 | * struct kmem_cache related prototypes | |
124 | */ | |
125 | void __init kmem_cache_init(void); | |
81819f0f | 126 | int slab_is_available(void); |
1da177e4 | 127 | |
2e892f43 | 128 | struct kmem_cache *kmem_cache_create(const char *, size_t, size_t, |
ebe29738 | 129 | unsigned long, |
51cc5068 | 130 | void (*)(void *)); |
2633d7a0 GC |
131 | struct kmem_cache * |
132 | kmem_cache_create_memcg(struct mem_cgroup *, const char *, size_t, size_t, | |
133 | unsigned long, void (*)(void *)); | |
2e892f43 CL |
134 | void kmem_cache_destroy(struct kmem_cache *); |
135 | int kmem_cache_shrink(struct kmem_cache *); | |
2e892f43 | 136 | void kmem_cache_free(struct kmem_cache *, void *); |
2e892f43 | 137 | |
0a31bd5f CL |
138 | /* |
139 | * Please use this macro to create slab caches. Simply specify the | |
140 | * name of the structure and maybe some flags that are listed above. | |
141 | * | |
142 | * The alignment of the struct determines object alignment. If you | |
143 | * f.e. add ____cacheline_aligned_in_smp to the struct declaration | |
144 | * then the objects will be properly aligned in SMP configurations. | |
145 | */ | |
146 | #define KMEM_CACHE(__struct, __flags) kmem_cache_create(#__struct,\ | |
147 | sizeof(struct __struct), __alignof__(struct __struct),\ | |
20c2df83 | 148 | (__flags), NULL) |
0a31bd5f | 149 | |
0aa817f0 CL |
150 | /* |
151 | * The largest kmalloc size supported by the slab allocators is | |
152 | * 32 megabyte (2^25) or the maximum allocatable page order if that is | |
153 | * less than 32 MB. | |
154 | * | |
155 | * WARNING: Its not easy to increase this value since the allocators have | |
156 | * to do various tricks to work around compiler limitations in order to | |
157 | * ensure proper constant folding. | |
158 | */ | |
debee076 CL |
159 | #define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \ |
160 | (MAX_ORDER + PAGE_SHIFT - 1) : 25) | |
0aa817f0 CL |
161 | |
162 | #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_HIGH) | |
163 | #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_HIGH - PAGE_SHIFT) | |
164 | ||
90810645 CL |
165 | /* |
166 | * Some archs want to perform DMA into kmalloc caches and need a guaranteed | |
167 | * alignment larger than the alignment of a 64-bit integer. | |
168 | * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that. | |
169 | */ | |
3192b920 CL |
170 | #ifdef ARCH_DMA_MINALIGN |
171 | #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN | |
172 | #else | |
173 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) | |
174 | #endif | |
175 | ||
90810645 CL |
176 | /* |
177 | * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment. | |
178 | * Intended for arches that get misalignment faults even for 64 bit integer | |
179 | * aligned buffers. | |
180 | */ | |
3192b920 CL |
181 | #ifndef ARCH_SLAB_MINALIGN |
182 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) | |
183 | #endif | |
ba6c496e GC |
184 | /* |
185 | * This is the main placeholder for memcg-related information in kmem caches. | |
186 | * struct kmem_cache will hold a pointer to it, so the memory cost while | |
187 | * disabled is 1 pointer. The runtime cost while enabled, gets bigger than it | |
188 | * would otherwise be if that would be bundled in kmem_cache: we'll need an | |
189 | * extra pointer chase. But the trade off clearly lays in favor of not | |
190 | * penalizing non-users. | |
191 | * | |
192 | * Both the root cache and the child caches will have it. For the root cache, | |
193 | * this will hold a dynamically allocated array large enough to hold | |
194 | * information about the currently limited memcgs in the system. | |
195 | * | |
196 | * Child caches will hold extra metadata needed for its operation. Fields are: | |
197 | * | |
198 | * @memcg: pointer to the memcg this cache belongs to | |
2633d7a0 GC |
199 | * @list: list_head for the list of all caches in this memcg |
200 | * @root_cache: pointer to the global, root cache, this cache was derived from | |
1f458cbf GC |
201 | * @dead: set to true after the memcg dies; the cache may still be around. |
202 | * @nr_pages: number of pages that belongs to this cache. | |
203 | * @destroy: worker to be called whenever we are ready, or believe we may be | |
204 | * ready, to destroy this cache. | |
ba6c496e GC |
205 | */ |
206 | struct memcg_cache_params { | |
207 | bool is_root_cache; | |
208 | union { | |
209 | struct kmem_cache *memcg_caches[0]; | |
2633d7a0 GC |
210 | struct { |
211 | struct mem_cgroup *memcg; | |
212 | struct list_head list; | |
213 | struct kmem_cache *root_cache; | |
1f458cbf GC |
214 | bool dead; |
215 | atomic_t nr_pages; | |
216 | struct work_struct destroy; | |
2633d7a0 | 217 | }; |
ba6c496e GC |
218 | }; |
219 | }; | |
220 | ||
2633d7a0 GC |
221 | int memcg_update_all_caches(int num_memcgs); |
222 | ||
2e892f43 CL |
223 | /* |
224 | * Common kmalloc functions provided by all allocators | |
225 | */ | |
93bc4e89 | 226 | void * __must_check __krealloc(const void *, size_t, gfp_t); |
fd76bab2 | 227 | void * __must_check krealloc(const void *, size_t, gfp_t); |
2e892f43 | 228 | void kfree(const void *); |
3ef0e5ba | 229 | void kzfree(const void *); |
fd76bab2 | 230 | size_t ksize(const void *); |
2e892f43 | 231 | |
81cda662 CL |
232 | /* |
233 | * Allocator specific definitions. These are mainly used to establish optimized | |
234 | * ways to convert kmalloc() calls to kmem_cache_alloc() invocations by | |
235 | * selecting the appropriate general cache at compile time. | |
236 | * | |
237 | * Allocators must define at least: | |
238 | * | |
239 | * kmem_cache_alloc() | |
240 | * __kmalloc() | |
241 | * kmalloc() | |
242 | * | |
243 | * Those wishing to support NUMA must also define: | |
244 | * | |
245 | * kmem_cache_alloc_node() | |
246 | * kmalloc_node() | |
247 | * | |
248 | * See each allocator definition file for additional comments and | |
249 | * implementation notes. | |
250 | */ | |
251 | #ifdef CONFIG_SLUB | |
252 | #include <linux/slub_def.h> | |
253 | #elif defined(CONFIG_SLOB) | |
254 | #include <linux/slob_def.h> | |
255 | #else | |
256 | #include <linux/slab_def.h> | |
257 | #endif | |
258 | ||
2e892f43 | 259 | /** |
a8203725 | 260 | * kmalloc_array - allocate memory for an array. |
2e892f43 CL |
261 | * @n: number of elements. |
262 | * @size: element size. | |
263 | * @flags: the type of memory to allocate. | |
800590f5 PD |
264 | * |
265 | * The @flags argument may be one of: | |
266 | * | |
267 | * %GFP_USER - Allocate memory on behalf of user. May sleep. | |
268 | * | |
269 | * %GFP_KERNEL - Allocate normal kernel ram. May sleep. | |
270 | * | |
6193a2ff | 271 | * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools. |
800590f5 PD |
272 | * For example, use this inside interrupt handlers. |
273 | * | |
274 | * %GFP_HIGHUSER - Allocate pages from high memory. | |
275 | * | |
276 | * %GFP_NOIO - Do not do any I/O at all while trying to get memory. | |
277 | * | |
278 | * %GFP_NOFS - Do not make any fs calls while trying to get memory. | |
279 | * | |
6193a2ff PM |
280 | * %GFP_NOWAIT - Allocation will not sleep. |
281 | * | |
282 | * %GFP_THISNODE - Allocate node-local memory only. | |
283 | * | |
284 | * %GFP_DMA - Allocation suitable for DMA. | |
285 | * Should only be used for kmalloc() caches. Otherwise, use a | |
286 | * slab created with SLAB_DMA. | |
287 | * | |
800590f5 PD |
288 | * Also it is possible to set different flags by OR'ing |
289 | * in one or more of the following additional @flags: | |
290 | * | |
291 | * %__GFP_COLD - Request cache-cold pages instead of | |
292 | * trying to return cache-warm pages. | |
293 | * | |
800590f5 PD |
294 | * %__GFP_HIGH - This allocation has high priority and may use emergency pools. |
295 | * | |
800590f5 PD |
296 | * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail |
297 | * (think twice before using). | |
298 | * | |
299 | * %__GFP_NORETRY - If memory is not immediately available, | |
300 | * then give up at once. | |
301 | * | |
302 | * %__GFP_NOWARN - If allocation fails, don't issue any warnings. | |
303 | * | |
304 | * %__GFP_REPEAT - If allocation fails initially, try once more before failing. | |
6193a2ff PM |
305 | * |
306 | * There are other flags available as well, but these are not intended | |
307 | * for general use, and so are not documented here. For a full list of | |
308 | * potential flags, always refer to linux/gfp.h. | |
800590f5 | 309 | */ |
a8203725 | 310 | static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags) |
1da177e4 | 311 | { |
a3860c1c | 312 | if (size != 0 && n > SIZE_MAX / size) |
6193a2ff | 313 | return NULL; |
a8203725 XW |
314 | return __kmalloc(n * size, flags); |
315 | } | |
316 | ||
317 | /** | |
318 | * kcalloc - allocate memory for an array. The memory is set to zero. | |
319 | * @n: number of elements. | |
320 | * @size: element size. | |
321 | * @flags: the type of memory to allocate (see kmalloc). | |
322 | */ | |
323 | static inline void *kcalloc(size_t n, size_t size, gfp_t flags) | |
324 | { | |
325 | return kmalloc_array(n, size, flags | __GFP_ZERO); | |
1da177e4 LT |
326 | } |
327 | ||
6193a2ff PM |
328 | #if !defined(CONFIG_NUMA) && !defined(CONFIG_SLOB) |
329 | /** | |
330 | * kmalloc_node - allocate memory from a specific node | |
331 | * @size: how many bytes of memory are required. | |
332 | * @flags: the type of memory to allocate (see kcalloc). | |
333 | * @node: node to allocate from. | |
334 | * | |
335 | * kmalloc() for non-local nodes, used to allocate from a specific node | |
336 | * if available. Equivalent to kmalloc() in the non-NUMA single-node | |
337 | * case. | |
338 | */ | |
55935a34 CL |
339 | static inline void *kmalloc_node(size_t size, gfp_t flags, int node) |
340 | { | |
341 | return kmalloc(size, flags); | |
342 | } | |
343 | ||
344 | static inline void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
345 | { | |
346 | return __kmalloc(size, flags); | |
347 | } | |
6193a2ff PM |
348 | |
349 | void *kmem_cache_alloc(struct kmem_cache *, gfp_t); | |
350 | ||
351 | static inline void *kmem_cache_alloc_node(struct kmem_cache *cachep, | |
352 | gfp_t flags, int node) | |
353 | { | |
354 | return kmem_cache_alloc(cachep, flags); | |
355 | } | |
356 | #endif /* !CONFIG_NUMA && !CONFIG_SLOB */ | |
55935a34 | 357 | |
1d2c8eea CH |
358 | /* |
359 | * kmalloc_track_caller is a special version of kmalloc that records the | |
360 | * calling function of the routine calling it for slab leak tracking instead | |
361 | * of just the calling function (confusing, eh?). | |
362 | * It's useful when the call to kmalloc comes from a widely-used standard | |
363 | * allocator where we care about the real place the memory allocation | |
364 | * request comes from. | |
365 | */ | |
7adde04a | 366 | #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_SLUB) || \ |
f3f74101 EG |
367 | (defined(CONFIG_SLAB) && defined(CONFIG_TRACING)) || \ |
368 | (defined(CONFIG_SLOB) && defined(CONFIG_TRACING)) | |
ce71e27c | 369 | extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long); |
1d2c8eea | 370 | #define kmalloc_track_caller(size, flags) \ |
ce71e27c | 371 | __kmalloc_track_caller(size, flags, _RET_IP_) |
2e892f43 CL |
372 | #else |
373 | #define kmalloc_track_caller(size, flags) \ | |
374 | __kmalloc(size, flags) | |
375 | #endif /* DEBUG_SLAB */ | |
1da177e4 | 376 | |
97e2bde4 | 377 | #ifdef CONFIG_NUMA |
8b98c169 CH |
378 | /* |
379 | * kmalloc_node_track_caller is a special version of kmalloc_node that | |
380 | * records the calling function of the routine calling it for slab leak | |
381 | * tracking instead of just the calling function (confusing, eh?). | |
382 | * It's useful when the call to kmalloc_node comes from a widely-used | |
383 | * standard allocator where we care about the real place the memory | |
384 | * allocation request comes from. | |
385 | */ | |
7adde04a | 386 | #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_SLUB) || \ |
f3f74101 EG |
387 | (defined(CONFIG_SLAB) && defined(CONFIG_TRACING)) || \ |
388 | (defined(CONFIG_SLOB) && defined(CONFIG_TRACING)) | |
ce71e27c | 389 | extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long); |
8b98c169 CH |
390 | #define kmalloc_node_track_caller(size, flags, node) \ |
391 | __kmalloc_node_track_caller(size, flags, node, \ | |
ce71e27c | 392 | _RET_IP_) |
2e892f43 CL |
393 | #else |
394 | #define kmalloc_node_track_caller(size, flags, node) \ | |
395 | __kmalloc_node(size, flags, node) | |
8b98c169 | 396 | #endif |
2e892f43 | 397 | |
8b98c169 | 398 | #else /* CONFIG_NUMA */ |
8b98c169 CH |
399 | |
400 | #define kmalloc_node_track_caller(size, flags, node) \ | |
401 | kmalloc_track_caller(size, flags) | |
97e2bde4 | 402 | |
dfcd3610 | 403 | #endif /* CONFIG_NUMA */ |
10cef602 | 404 | |
81cda662 CL |
405 | /* |
406 | * Shortcuts | |
407 | */ | |
408 | static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags) | |
409 | { | |
410 | return kmem_cache_alloc(k, flags | __GFP_ZERO); | |
411 | } | |
412 | ||
413 | /** | |
414 | * kzalloc - allocate memory. The memory is set to zero. | |
415 | * @size: how many bytes of memory are required. | |
416 | * @flags: the type of memory to allocate (see kmalloc). | |
417 | */ | |
418 | static inline void *kzalloc(size_t size, gfp_t flags) | |
419 | { | |
420 | return kmalloc(size, flags | __GFP_ZERO); | |
421 | } | |
422 | ||
979b0fea JL |
423 | /** |
424 | * kzalloc_node - allocate zeroed memory from a particular memory node. | |
425 | * @size: how many bytes of memory are required. | |
426 | * @flags: the type of memory to allocate (see kmalloc). | |
427 | * @node: memory node from which to allocate | |
428 | */ | |
429 | static inline void *kzalloc_node(size_t size, gfp_t flags, int node) | |
430 | { | |
431 | return kmalloc_node(size, flags | __GFP_ZERO, node); | |
432 | } | |
433 | ||
242860a4 EG |
434 | /* |
435 | * Determine the size of a slab object | |
436 | */ | |
437 | static inline unsigned int kmem_cache_size(struct kmem_cache *s) | |
438 | { | |
439 | return s->object_size; | |
440 | } | |
441 | ||
7e85ee0c PE |
442 | void __init kmem_cache_init_late(void); |
443 | ||
1da177e4 | 444 | #endif /* _LINUX_SLAB_H */ |