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,
119 kmem_cache_create_memcg(struct mem_cgroup *, const char *, size_t, size_t,
120 unsigned long, void (*)(void *), 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 allocates up to order 2 pages directly and otherwise
209 * passes the request 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 page size and larger requests 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_MAX 30
225 #define KMALLOC_SHIFT_HIGH PAGE_SHIFT
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 extern struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
247 #ifdef CONFIG_ZONE_DMA
248 extern struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
252 * Figure out which kmalloc slab an allocation of a certain size
256 * 2 = 120 .. 192 bytes
257 * n = 2^(n-1) .. 2^n -1
259 static __always_inline int kmalloc_index(size_t size)
264 if (size <= KMALLOC_MIN_SIZE)
265 return KMALLOC_SHIFT_LOW;
267 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
269 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
271 if (size <= 8) return 3;
272 if (size <= 16) return 4;
273 if (size <= 32) return 5;
274 if (size <= 64) return 6;
275 if (size <= 128) return 7;
276 if (size <= 256) return 8;
277 if (size <= 512) return 9;
278 if (size <= 1024) return 10;
279 if (size <= 2 * 1024) return 11;
280 if (size <= 4 * 1024) return 12;
281 if (size <= 8 * 1024) return 13;
282 if (size <= 16 * 1024) return 14;
283 if (size <= 32 * 1024) return 15;
284 if (size <= 64 * 1024) return 16;
285 if (size <= 128 * 1024) return 17;
286 if (size <= 256 * 1024) return 18;
287 if (size <= 512 * 1024) return 19;
288 if (size <= 1024 * 1024) return 20;
289 if (size <= 2 * 1024 * 1024) return 21;
290 if (size <= 4 * 1024 * 1024) return 22;
291 if (size <= 8 * 1024 * 1024) return 23;
292 if (size <= 16 * 1024 * 1024) return 24;
293 if (size <= 32 * 1024 * 1024) return 25;
294 if (size <= 64 * 1024 * 1024) return 26;
297 /* Will never be reached. Needed because the compiler may complain */
300 #endif /* !CONFIG_SLOB */
302 void *__kmalloc(size_t size, gfp_t flags);
303 void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags);
306 void *__kmalloc_node(size_t size, gfp_t flags, int node);
307 void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node);
309 static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
311 return __kmalloc(size, flags);
314 static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
316 return kmem_cache_alloc(s, flags);
320 #ifdef CONFIG_TRACING
321 extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t);
324 extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
326 int node, size_t size);
328 static __always_inline void *
329 kmem_cache_alloc_node_trace(struct kmem_cache *s,
331 int node, size_t size)
333 return kmem_cache_alloc_trace(s, gfpflags, size);
335 #endif /* CONFIG_NUMA */
337 #else /* CONFIG_TRACING */
338 static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
339 gfp_t flags, size_t size)
341 return kmem_cache_alloc(s, flags);
344 static __always_inline void *
345 kmem_cache_alloc_node_trace(struct kmem_cache *s,
347 int node, size_t size)
349 return kmem_cache_alloc_node(s, gfpflags, node);
351 #endif /* CONFIG_TRACING */
354 #include <linux/slab_def.h>
358 #include <linux/slub_def.h>
361 static __always_inline void *
362 kmalloc_order(size_t size, gfp_t flags, unsigned int order)
366 flags |= (__GFP_COMP | __GFP_KMEMCG);
367 ret = (void *) __get_free_pages(flags, order);
368 kmemleak_alloc(ret, size, 1, flags);
372 #ifdef CONFIG_TRACING
373 extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order);
375 static __always_inline void *
376 kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
378 return kmalloc_order(size, flags, order);
382 static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
384 unsigned int order = get_order(size);
385 return kmalloc_order_trace(size, flags, order);
389 * kmalloc - allocate memory
390 * @size: how many bytes of memory are required.
391 * @flags: the type of memory to allocate.
393 * kmalloc is the normal method of allocating memory
394 * for objects smaller than page size in the kernel.
396 * The @flags argument may be one of:
398 * %GFP_USER - Allocate memory on behalf of user. May sleep.
400 * %GFP_KERNEL - Allocate normal kernel ram. May sleep.
402 * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools.
403 * For example, use this inside interrupt handlers.
405 * %GFP_HIGHUSER - Allocate pages from high memory.
407 * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
409 * %GFP_NOFS - Do not make any fs calls while trying to get memory.
411 * %GFP_NOWAIT - Allocation will not sleep.
413 * %GFP_THISNODE - Allocate node-local memory only.
415 * %GFP_DMA - Allocation suitable for DMA.
416 * Should only be used for kmalloc() caches. Otherwise, use a
417 * slab created with SLAB_DMA.
419 * Also it is possible to set different flags by OR'ing
420 * in one or more of the following additional @flags:
422 * %__GFP_COLD - Request cache-cold pages instead of
423 * trying to return cache-warm pages.
425 * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
427 * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
428 * (think twice before using).
430 * %__GFP_NORETRY - If memory is not immediately available,
431 * then give up at once.
433 * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
435 * %__GFP_REPEAT - If allocation fails initially, try once more before failing.
437 * There are other flags available as well, but these are not intended
438 * for general use, and so are not documented here. For a full list of
439 * potential flags, always refer to linux/gfp.h.
441 static __always_inline void *kmalloc(size_t size, gfp_t flags)
443 if (__builtin_constant_p(size)) {
444 if (size > KMALLOC_MAX_CACHE_SIZE)
445 return kmalloc_large(size, flags);
447 if (!(flags & GFP_DMA)) {
448 int index = kmalloc_index(size);
451 return ZERO_SIZE_PTR;
453 return kmem_cache_alloc_trace(kmalloc_caches[index],
458 return __kmalloc(size, flags);
462 * Determine size used for the nth kmalloc cache.
463 * return size or 0 if a kmalloc cache for that
464 * size does not exist
466 static __always_inline int kmalloc_size(int n)
472 if (n == 1 && KMALLOC_MIN_SIZE <= 32)
475 if (n == 2 && KMALLOC_MIN_SIZE <= 64)
481 static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
484 if (__builtin_constant_p(size) &&
485 size <= KMALLOC_MAX_CACHE_SIZE && !(flags & GFP_DMA)) {
486 int i = kmalloc_index(size);
489 return ZERO_SIZE_PTR;
491 return kmem_cache_alloc_node_trace(kmalloc_caches[i],
495 return __kmalloc_node(size, flags, node);
499 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
500 * Intended for arches that get misalignment faults even for 64 bit integer
503 #ifndef ARCH_SLAB_MINALIGN
504 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
507 * This is the main placeholder for memcg-related information in kmem caches.
508 * struct kmem_cache will hold a pointer to it, so the memory cost while
509 * disabled is 1 pointer. The runtime cost while enabled, gets bigger than it
510 * would otherwise be if that would be bundled in kmem_cache: we'll need an
511 * extra pointer chase. But the trade off clearly lays in favor of not
512 * penalizing non-users.
514 * Both the root cache and the child caches will have it. For the root cache,
515 * this will hold a dynamically allocated array large enough to hold
516 * information about the currently limited memcgs in the system. To allow the
517 * array to be accessed without taking any locks, on relocation we free the old
518 * version only after a grace period.
520 * Child caches will hold extra metadata needed for its operation. Fields are:
522 * @memcg: pointer to the memcg this cache belongs to
523 * @list: list_head for the list of all caches in this memcg
524 * @root_cache: pointer to the global, root cache, this cache was derived from
525 * @dead: set to true after the memcg dies; the cache may still be around.
526 * @nr_pages: number of pages that belongs to this cache.
527 * @destroy: worker to be called whenever we are ready, or believe we may be
528 * ready, to destroy this cache.
530 struct memcg_cache_params {
534 struct rcu_head rcu_head;
535 struct kmem_cache *memcg_caches[0];
538 struct mem_cgroup *memcg;
539 struct list_head list;
540 struct kmem_cache *root_cache;
543 struct work_struct destroy;
548 int memcg_update_all_caches(int num_memcgs);
551 int cache_show(struct kmem_cache *s, struct seq_file *m);
552 void print_slabinfo_header(struct seq_file *m);
555 * kmalloc_array - allocate memory for an array.
556 * @n: number of elements.
557 * @size: element size.
558 * @flags: the type of memory to allocate (see kmalloc).
560 static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
562 if (size != 0 && n > SIZE_MAX / size)
564 return __kmalloc(n * size, flags);
568 * kcalloc - allocate memory for an array. The memory is set to zero.
569 * @n: number of elements.
570 * @size: element size.
571 * @flags: the type of memory to allocate (see kmalloc).
573 static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
575 return kmalloc_array(n, size, flags | __GFP_ZERO);
579 * kmalloc_track_caller is a special version of kmalloc that records the
580 * calling function of the routine calling it for slab leak tracking instead
581 * of just the calling function (confusing, eh?).
582 * It's useful when the call to kmalloc comes from a widely-used standard
583 * allocator where we care about the real place the memory allocation
584 * request comes from.
586 #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_SLUB) || \
587 (defined(CONFIG_SLAB) && defined(CONFIG_TRACING)) || \
588 (defined(CONFIG_SLOB) && defined(CONFIG_TRACING))
589 extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
590 #define kmalloc_track_caller(size, flags) \
591 __kmalloc_track_caller(size, flags, _RET_IP_)
593 #define kmalloc_track_caller(size, flags) \
594 __kmalloc(size, flags)
595 #endif /* DEBUG_SLAB */
599 * kmalloc_node_track_caller is a special version of kmalloc_node that
600 * records the calling function of the routine calling it for slab leak
601 * tracking instead of just the calling function (confusing, eh?).
602 * It's useful when the call to kmalloc_node comes from a widely-used
603 * standard allocator where we care about the real place the memory
604 * allocation request comes from.
606 #if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_SLUB) || \
607 (defined(CONFIG_SLAB) && defined(CONFIG_TRACING)) || \
608 (defined(CONFIG_SLOB) && defined(CONFIG_TRACING))
609 extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
610 #define kmalloc_node_track_caller(size, flags, node) \
611 __kmalloc_node_track_caller(size, flags, node, \
614 #define kmalloc_node_track_caller(size, flags, node) \
615 __kmalloc_node(size, flags, node)
618 #else /* CONFIG_NUMA */
620 #define kmalloc_node_track_caller(size, flags, node) \
621 kmalloc_track_caller(size, flags)
623 #endif /* CONFIG_NUMA */
628 static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
630 return kmem_cache_alloc(k, flags | __GFP_ZERO);
634 * kzalloc - allocate memory. The memory is set to zero.
635 * @size: how many bytes of memory are required.
636 * @flags: the type of memory to allocate (see kmalloc).
638 static inline void *kzalloc(size_t size, gfp_t flags)
640 return kmalloc(size, flags | __GFP_ZERO);
644 * kzalloc_node - allocate zeroed memory from a particular memory node.
645 * @size: how many bytes of memory are required.
646 * @flags: the type of memory to allocate (see kmalloc).
647 * @node: memory node from which to allocate
649 static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
651 return kmalloc_node(size, flags | __GFP_ZERO, node);
655 * Determine the size of a slab object
657 static inline unsigned int kmem_cache_size(struct kmem_cache *s)
659 return s->object_size;
662 void __init kmem_cache_init_late(void);
664 #endif /* _LINUX_SLAB_H */