2 * Written by Mark Hemment, 1996 (markhe@nextd.demon.co.uk).
4 * (C) SGI 2006, Christoph Lameter
5 * Cleaned up and restructured to ease the addition of alternative
6 * implementations of SLAB allocators.
7 * (C) Linux Foundation 2008-2013
8 * Unified interface for all slab allocators
14 #include <linux/gfp.h>
15 #include <linux/types.h>
16 #include <linux/workqueue.h>
20 * Flags to pass to kmem_cache_create().
21 * The ones marked DEBUG are only valid if CONFIG_SLAB_DEBUG is set.
23 #define SLAB_DEBUG_FREE 0x00000100UL /* DEBUG: Perform (expensive) checks on free */
24 #define SLAB_RED_ZONE 0x00000400UL /* DEBUG: Red zone objs in a cache */
25 #define SLAB_POISON 0x00000800UL /* DEBUG: Poison objects */
26 #define SLAB_HWCACHE_ALIGN 0x00002000UL /* Align objs on cache lines */
27 #define SLAB_CACHE_DMA 0x00004000UL /* Use GFP_DMA memory */
28 #define SLAB_STORE_USER 0x00010000UL /* DEBUG: Store the last owner for bug hunting */
29 #define SLAB_PANIC 0x00040000UL /* Panic if kmem_cache_create() fails */
31 * SLAB_DESTROY_BY_RCU - **WARNING** READ THIS!
33 * This delays freeing the SLAB page by a grace period, it does _NOT_
34 * delay object freeing. This means that if you do kmem_cache_free()
35 * that memory location is free to be reused at any time. Thus it may
36 * be possible to see another object there in the same RCU grace period.
38 * This feature only ensures the memory location backing the object
39 * stays valid, the trick to using this is relying on an independent
40 * object validation pass. Something like:
44 * obj = lockless_lookup(key);
46 * if (!try_get_ref(obj)) // might fail for free objects
49 * if (obj->key != key) { // not the object we expected
56 * This is useful if we need to approach a kernel structure obliquely,
57 * from its address obtained without the usual locking. We can lock
58 * the structure to stabilize it and check it's still at the given address,
59 * only if we can be sure that the memory has not been meanwhile reused
60 * for some other kind of object (which our subsystem's lock might corrupt).
62 * rcu_read_lock before reading the address, then rcu_read_unlock after
63 * taking the spinlock within the structure expected at that address.
65 #define SLAB_DESTROY_BY_RCU 0x00080000UL /* Defer freeing slabs to RCU */
66 #define SLAB_MEM_SPREAD 0x00100000UL /* Spread some memory over cpuset */
67 #define SLAB_TRACE 0x00200000UL /* Trace allocations and frees */
69 /* Flag to prevent checks on free */
70 #ifdef CONFIG_DEBUG_OBJECTS
71 # define SLAB_DEBUG_OBJECTS 0x00400000UL
73 # define SLAB_DEBUG_OBJECTS 0x00000000UL
76 #define SLAB_NOLEAKTRACE 0x00800000UL /* Avoid kmemleak tracing */
78 /* Don't track use of uninitialized memory */
79 #ifdef CONFIG_KMEMCHECK
80 # define SLAB_NOTRACK 0x01000000UL
82 # define SLAB_NOTRACK 0x00000000UL
84 #ifdef CONFIG_FAILSLAB
85 # define SLAB_FAILSLAB 0x02000000UL /* Fault injection mark */
87 # define SLAB_FAILSLAB 0x00000000UL
90 /* The following flags affect the page allocator grouping pages by mobility */
91 #define SLAB_RECLAIM_ACCOUNT 0x00020000UL /* Objects are reclaimable */
92 #define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
94 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
96 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
98 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
99 * Both make kfree a no-op.
101 #define ZERO_SIZE_PTR ((void *)16)
103 #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
104 (unsigned long)ZERO_SIZE_PTR)
106 #include <linux/kmemleak.h>
110 * struct kmem_cache related prototypes
112 void __init
kmem_cache_init(void);
113 int slab_is_available(void);
115 struct kmem_cache
*kmem_cache_create(const char *, size_t, size_t,
118 #ifdef CONFIG_MEMCG_KMEM
119 void kmem_cache_create_memcg(struct mem_cgroup
*, struct kmem_cache
*);
121 void kmem_cache_destroy(struct kmem_cache
*);
122 int kmem_cache_shrink(struct kmem_cache
*);
123 void kmem_cache_free(struct kmem_cache
*, void *);
126 * Please use this macro to create slab caches. Simply specify the
127 * name of the structure and maybe some flags that are listed above.
129 * The alignment of the struct determines object alignment. If you
130 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
131 * then the objects will be properly aligned in SMP configurations.
133 #define KMEM_CACHE(__struct, __flags) kmem_cache_create(#__struct,\
134 sizeof(struct __struct), __alignof__(struct __struct),\
138 * Common kmalloc functions provided by all allocators
140 void * __must_check
__krealloc(const void *, size_t, gfp_t
);
141 void * __must_check
krealloc(const void *, size_t, gfp_t
);
142 void kfree(const void *);
143 void kzfree(const void *);
144 size_t ksize(const void *);
147 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
148 * alignment larger than the alignment of a 64-bit integer.
149 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
151 #if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
152 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
153 #define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
154 #define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
156 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
161 * Common fields provided in kmem_cache by all slab allocators
162 * This struct is either used directly by the allocator (SLOB)
163 * or the allocator must include definitions for all fields
164 * provided in kmem_cache_common in their definition of kmem_cache.
166 * Once we can do anonymous structs (C11 standard) we could put a
167 * anonymous struct definition in these allocators so that the
168 * separate allocations in the kmem_cache structure of SLAB and
169 * SLUB is no longer needed.
172 unsigned int object_size
;/* The original size of the object */
173 unsigned int size
; /* The aligned/padded/added on size */
174 unsigned int align
; /* Alignment as calculated */
175 unsigned long flags
; /* Active flags on the slab */
176 const char *name
; /* Slab name for sysfs */
177 int refcount
; /* Use counter */
178 void (*ctor
)(void *); /* Called on object slot creation */
179 struct list_head list
; /* List of all slab caches on the system */
182 #endif /* CONFIG_SLOB */
185 * Kmalloc array related definitions
190 * The largest kmalloc size supported by the SLAB allocators is
191 * 32 megabyte (2^25) or the maximum allocatable page order if that is
194 * WARNING: Its not easy to increase this value since the allocators have
195 * to do various tricks to work around compiler limitations in order to
196 * ensure proper constant folding.
198 #define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
199 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
200 #define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
201 #ifndef KMALLOC_SHIFT_LOW
202 #define KMALLOC_SHIFT_LOW 5
208 * SLUB directly allocates requests fitting in to an order-1 page
209 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
211 #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
212 #define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT)
213 #ifndef KMALLOC_SHIFT_LOW
214 #define KMALLOC_SHIFT_LOW 3
220 * SLOB passes all requests larger than one page to the page allocator.
221 * No kmalloc array is necessary since objects of different sizes can
222 * be allocated from the same page.
224 #define KMALLOC_SHIFT_HIGH PAGE_SHIFT
225 #define KMALLOC_SHIFT_MAX 30
226 #ifndef KMALLOC_SHIFT_LOW
227 #define KMALLOC_SHIFT_LOW 3
231 /* Maximum allocatable size */
232 #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
233 /* Maximum size for which we actually use a slab cache */
234 #define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
235 /* Maximum order allocatable via the slab allocagtor */
236 #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
241 #ifndef KMALLOC_MIN_SIZE
242 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
246 * This restriction comes from byte sized index implementation.
247 * Page size is normally 2^12 bytes and, in this case, if we want to use
248 * byte sized index which can represent 2^8 entries, the size of the object
249 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
250 * If minimum size of kmalloc is less than 16, we use it as minimum object
251 * size and give up to use byte sized index.
253 #define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
254 (KMALLOC_MIN_SIZE) : 16)
257 extern struct kmem_cache
*kmalloc_caches
[KMALLOC_SHIFT_HIGH
+ 1];
258 #ifdef CONFIG_ZONE_DMA
259 extern struct kmem_cache
*kmalloc_dma_caches
[KMALLOC_SHIFT_HIGH
+ 1];
263 * Figure out which kmalloc slab an allocation of a certain size
267 * 2 = 120 .. 192 bytes
268 * n = 2^(n-1) .. 2^n -1
270 static __always_inline
int kmalloc_index(size_t size
)
275 if (size
<= KMALLOC_MIN_SIZE
)
276 return KMALLOC_SHIFT_LOW
;
278 if (KMALLOC_MIN_SIZE
<= 32 && size
> 64 && size
<= 96)
280 if (KMALLOC_MIN_SIZE
<= 64 && size
> 128 && size
<= 192)
282 if (size
<= 8) return 3;
283 if (size
<= 16) return 4;
284 if (size
<= 32) return 5;
285 if (size
<= 64) return 6;
286 if (size
<= 128) return 7;
287 if (size
<= 256) return 8;
288 if (size
<= 512) return 9;
289 if (size
<= 1024) return 10;
290 if (size
<= 2 * 1024) return 11;
291 if (size
<= 4 * 1024) return 12;
292 if (size
<= 8 * 1024) return 13;
293 if (size
<= 16 * 1024) return 14;
294 if (size
<= 32 * 1024) return 15;
295 if (size
<= 64 * 1024) return 16;
296 if (size
<= 128 * 1024) return 17;
297 if (size
<= 256 * 1024) return 18;
298 if (size
<= 512 * 1024) return 19;
299 if (size
<= 1024 * 1024) return 20;
300 if (size
<= 2 * 1024 * 1024) return 21;
301 if (size
<= 4 * 1024 * 1024) return 22;
302 if (size
<= 8 * 1024 * 1024) return 23;
303 if (size
<= 16 * 1024 * 1024) return 24;
304 if (size
<= 32 * 1024 * 1024) return 25;
305 if (size
<= 64 * 1024 * 1024) return 26;
308 /* Will never be reached. Needed because the compiler may complain */
311 #endif /* !CONFIG_SLOB */
313 void *__kmalloc(size_t size
, gfp_t flags
);
314 void *kmem_cache_alloc(struct kmem_cache
*, gfp_t flags
);
317 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
);
318 void *kmem_cache_alloc_node(struct kmem_cache
*, gfp_t flags
, int node
);
320 static __always_inline
void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
322 return __kmalloc(size
, flags
);
325 static __always_inline
void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t flags
, int node
)
327 return kmem_cache_alloc(s
, flags
);
331 #ifdef CONFIG_TRACING
332 extern void *kmem_cache_alloc_trace(struct kmem_cache
*, gfp_t
, size_t);
335 extern void *kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
337 int node
, size_t size
);
339 static __always_inline
void *
340 kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
342 int node
, size_t size
)
344 return kmem_cache_alloc_trace(s
, gfpflags
, size
);
346 #endif /* CONFIG_NUMA */
348 #else /* CONFIG_TRACING */
349 static __always_inline
void *kmem_cache_alloc_trace(struct kmem_cache
*s
,
350 gfp_t flags
, size_t size
)
352 return kmem_cache_alloc(s
, flags
);
355 static __always_inline
void *
356 kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
358 int node
, size_t size
)
360 return kmem_cache_alloc_node(s
, gfpflags
, node
);
362 #endif /* CONFIG_TRACING */
365 #include <linux/slab_def.h>
369 #include <linux/slub_def.h>
372 static __always_inline
void *
373 kmalloc_order(size_t size
, gfp_t flags
, unsigned int order
)
377 flags
|= (__GFP_COMP
| __GFP_KMEMCG
);
378 ret
= (void *) __get_free_pages(flags
, order
);
379 kmemleak_alloc(ret
, size
, 1, flags
);
383 #ifdef CONFIG_TRACING
384 extern void *kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
);
386 static __always_inline
void *
387 kmalloc_order_trace(size_t size
, gfp_t flags
, unsigned int order
)
389 return kmalloc_order(size
, flags
, order
);
393 static __always_inline
void *kmalloc_large(size_t size
, gfp_t flags
)
395 unsigned int order
= get_order(size
);
396 return kmalloc_order_trace(size
, flags
, order
);
400 * kmalloc - allocate memory
401 * @size: how many bytes of memory are required.
402 * @flags: the type of memory to allocate.
404 * kmalloc is the normal method of allocating memory
405 * for objects smaller than page size in the kernel.
407 * The @flags argument may be one of:
409 * %GFP_USER - Allocate memory on behalf of user. May sleep.
411 * %GFP_KERNEL - Allocate normal kernel ram. May sleep.
413 * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools.
414 * For example, use this inside interrupt handlers.
416 * %GFP_HIGHUSER - Allocate pages from high memory.
418 * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
420 * %GFP_NOFS - Do not make any fs calls while trying to get memory.
422 * %GFP_NOWAIT - Allocation will not sleep.
424 * %__GFP_THISNODE - Allocate node-local memory only.
426 * %GFP_DMA - Allocation suitable for DMA.
427 * Should only be used for kmalloc() caches. Otherwise, use a
428 * slab created with SLAB_DMA.
430 * Also it is possible to set different flags by OR'ing
431 * in one or more of the following additional @flags:
433 * %__GFP_COLD - Request cache-cold pages instead of
434 * trying to return cache-warm pages.
436 * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
438 * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
439 * (think twice before using).
441 * %__GFP_NORETRY - If memory is not immediately available,
442 * then give up at once.
444 * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
446 * %__GFP_REPEAT - If allocation fails initially, try once more before failing.
448 * There are other flags available as well, but these are not intended
449 * for general use, and so are not documented here. For a full list of
450 * potential flags, always refer to linux/gfp.h.
452 static __always_inline
void *kmalloc(size_t size
, gfp_t flags
)
454 if (__builtin_constant_p(size
)) {
455 if (size
> KMALLOC_MAX_CACHE_SIZE
)
456 return kmalloc_large(size
, flags
);
458 if (!(flags
& GFP_DMA
)) {
459 int index
= kmalloc_index(size
);
462 return ZERO_SIZE_PTR
;
464 return kmem_cache_alloc_trace(kmalloc_caches
[index
],
469 return __kmalloc(size
, flags
);
473 * Determine size used for the nth kmalloc cache.
474 * return size or 0 if a kmalloc cache for that
475 * size does not exist
477 static __always_inline
int kmalloc_size(int n
)
483 if (n
== 1 && KMALLOC_MIN_SIZE
<= 32)
486 if (n
== 2 && KMALLOC_MIN_SIZE
<= 64)
492 static __always_inline
void *kmalloc_node(size_t size
, gfp_t flags
, int node
)
495 if (__builtin_constant_p(size
) &&
496 size
<= KMALLOC_MAX_CACHE_SIZE
&& !(flags
& GFP_DMA
)) {
497 int i
= kmalloc_index(size
);
500 return ZERO_SIZE_PTR
;
502 return kmem_cache_alloc_node_trace(kmalloc_caches
[i
],
506 return __kmalloc_node(size
, flags
, node
);
510 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
511 * Intended for arches that get misalignment faults even for 64 bit integer
514 #ifndef ARCH_SLAB_MINALIGN
515 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
518 * This is the main placeholder for memcg-related information in kmem caches.
519 * struct kmem_cache will hold a pointer to it, so the memory cost while
520 * disabled is 1 pointer. The runtime cost while enabled, gets bigger than it
521 * would otherwise be if that would be bundled in kmem_cache: we'll need an
522 * extra pointer chase. But the trade off clearly lays in favor of not
523 * penalizing non-users.
525 * Both the root cache and the child caches will have it. For the root cache,
526 * this will hold a dynamically allocated array large enough to hold
527 * information about the currently limited memcgs in the system. To allow the
528 * array to be accessed without taking any locks, on relocation we free the old
529 * version only after a grace period.
531 * Child caches will hold extra metadata needed for its operation. Fields are:
533 * @memcg: pointer to the memcg this cache belongs to
534 * @list: list_head for the list of all caches in this memcg
535 * @root_cache: pointer to the global, root cache, this cache was derived from
536 * @dead: set to true after the memcg dies; the cache may still be around.
537 * @nr_pages: number of pages that belongs to this cache.
538 * @destroy: worker to be called whenever we are ready, or believe we may be
539 * ready, to destroy this cache.
541 struct memcg_cache_params
{
545 struct rcu_head rcu_head
;
546 struct kmem_cache
*memcg_caches
[0];
549 struct mem_cgroup
*memcg
;
550 struct list_head list
;
551 struct kmem_cache
*root_cache
;
554 struct work_struct destroy
;
559 int memcg_update_all_caches(int num_memcgs
);
562 int cache_show(struct kmem_cache
*s
, struct seq_file
*m
);
563 void print_slabinfo_header(struct seq_file
*m
);
566 * kmalloc_array - allocate memory for an array.
567 * @n: number of elements.
568 * @size: element size.
569 * @flags: the type of memory to allocate (see kmalloc).
571 static inline void *kmalloc_array(size_t n
, size_t size
, gfp_t flags
)
573 if (size
!= 0 && n
> SIZE_MAX
/ size
)
575 return __kmalloc(n
* size
, flags
);
579 * kcalloc - allocate memory for an array. The memory is set to zero.
580 * @n: number of elements.
581 * @size: element size.
582 * @flags: the type of memory to allocate (see kmalloc).
584 static inline void *kcalloc(size_t n
, size_t size
, gfp_t flags
)
586 return kmalloc_array(n
, size
, flags
| __GFP_ZERO
);
590 * kmalloc_track_caller is a special version of kmalloc that records the
591 * calling function of the routine calling it for slab leak tracking instead
592 * of just the calling function (confusing, eh?).
593 * It's useful when the call to kmalloc comes from a widely-used standard
594 * allocator where we care about the real place the memory allocation
595 * request comes from.
597 #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_SLUB) || \
598 (defined(CONFIG_SLAB) && defined(CONFIG_TRACING)) || \
599 (defined(CONFIG_SLOB) && defined(CONFIG_TRACING))
600 extern void *__kmalloc_track_caller(size_t, gfp_t
, unsigned long);
601 #define kmalloc_track_caller(size, flags) \
602 __kmalloc_track_caller(size, flags, _RET_IP_)
604 #define kmalloc_track_caller(size, flags) \
605 __kmalloc(size, flags)
606 #endif /* DEBUG_SLAB */
610 * kmalloc_node_track_caller is a special version of kmalloc_node that
611 * records the calling function of the routine calling it for slab leak
612 * tracking instead of just the calling function (confusing, eh?).
613 * It's useful when the call to kmalloc_node comes from a widely-used
614 * standard allocator where we care about the real place the memory
615 * allocation request comes from.
617 #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_SLUB) || \
618 (defined(CONFIG_SLAB) && defined(CONFIG_TRACING)) || \
619 (defined(CONFIG_SLOB) && defined(CONFIG_TRACING))
620 extern void *__kmalloc_node_track_caller(size_t, gfp_t
, int, unsigned long);
621 #define kmalloc_node_track_caller(size, flags, node) \
622 __kmalloc_node_track_caller(size, flags, node, \
625 #define kmalloc_node_track_caller(size, flags, node) \
626 __kmalloc_node(size, flags, node)
629 #else /* CONFIG_NUMA */
631 #define kmalloc_node_track_caller(size, flags, node) \
632 kmalloc_track_caller(size, flags)
634 #endif /* CONFIG_NUMA */
639 static inline void *kmem_cache_zalloc(struct kmem_cache
*k
, gfp_t flags
)
641 return kmem_cache_alloc(k
, flags
| __GFP_ZERO
);
645 * kzalloc - allocate memory. The memory is set to zero.
646 * @size: how many bytes of memory are required.
647 * @flags: the type of memory to allocate (see kmalloc).
649 static inline void *kzalloc(size_t size
, gfp_t flags
)
651 return kmalloc(size
, flags
| __GFP_ZERO
);
655 * kzalloc_node - allocate zeroed memory from a particular memory node.
656 * @size: how many bytes of memory are required.
657 * @flags: the type of memory to allocate (see kmalloc).
658 * @node: memory node from which to allocate
660 static inline void *kzalloc_node(size_t size
, gfp_t flags
, int node
)
662 return kmalloc_node(size
, flags
| __GFP_ZERO
, node
);
666 * Determine the size of a slab object
668 static inline unsigned int kmem_cache_size(struct kmem_cache
*s
)
670 return s
->object_size
;
673 void __init
kmem_cache_init_late(void);
675 #endif /* _LINUX_SLAB_H */
This page took 0.048049 seconds and 5 git commands to generate.