blob: 084b12bad198232426c6beb27cd348d72797b344 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Christoph Lameter2e892f42006-12-13 00:34:23 -08002 * Written by Mark Hemment, 1996 (markhe@nextd.demon.co.uk).
3 *
Christoph Lametercde53532008-07-04 09:59:22 -07004 * (C) SGI 2006, Christoph Lameter
Christoph Lameter2e892f42006-12-13 00:34:23 -08005 * Cleaned up and restructured to ease the addition of alternative
6 * implementations of SLAB allocators.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +00007 * (C) Linux Foundation 2008-2013
8 * Unified interface for all slab allocators
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 */
10
11#ifndef _LINUX_SLAB_H
12#define _LINUX_SLAB_H
13
Andrew Morton1b1cec42006-12-06 20:33:22 -080014#include <linux/gfp.h>
Andrew Morton1b1cec42006-12-06 20:33:22 -080015#include <linux/types.h>
Glauber Costa1f458cb2012-12-18 14:22:50 -080016#include <linux/workqueue.h>
17
Linus Torvalds1da177e2005-04-16 15:20:36 -070018
Christoph Lameter2e892f42006-12-13 00:34:23 -080019/*
20 * Flags to pass to kmem_cache_create().
David Rientjes124dee02015-04-14 15:44:28 -070021 * The ones marked DEBUG are only valid if CONFIG_DEBUG_SLAB is set.
Linus Torvalds1da177e2005-04-16 15:20:36 -070022 */
Laura Abbottbecfda62016-03-15 14:55:06 -070023#define SLAB_CONSISTENCY_CHECKS 0x00000100UL /* DEBUG: Perform (expensive) checks on alloc/free */
Christoph Lameter55935a32006-12-13 00:34:24 -080024#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 */
Christoph Lameter2e892f42006-12-13 00:34:23 -080027#define SLAB_CACHE_DMA 0x00004000UL /* Use GFP_DMA memory */
Christoph Lameter2e892f42006-12-13 00:34:23 -080028#define SLAB_STORE_USER 0x00010000UL /* DEBUG: Store the last owner for bug hunting */
Christoph Lameter2e892f42006-12-13 00:34:23 -080029#define SLAB_PANIC 0x00040000UL /* Panic if kmem_cache_create() fails */
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020030/*
31 * SLAB_DESTROY_BY_RCU - **WARNING** READ THIS!
32 *
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.
37 *
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:
41 *
42 * rcu_read_lock()
43 * again:
44 * obj = lockless_lookup(key);
45 * if (obj) {
46 * if (!try_get_ref(obj)) // might fail for free objects
47 * goto again;
48 *
49 * if (obj->key != key) { // not the object we expected
50 * put_ref(obj);
51 * goto again;
52 * }
53 * }
54 * rcu_read_unlock();
55 *
Joonsoo Kim68126702013-10-24 10:07:42 +090056 * 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).
61 *
62 * rcu_read_lock before reading the address, then rcu_read_unlock after
63 * taking the spinlock within the structure expected at that address.
Peter Zijlstrad7de4c12008-11-13 20:40:12 +020064 */
Christoph Lameter2e892f42006-12-13 00:34:23 -080065#define SLAB_DESTROY_BY_RCU 0x00080000UL /* Defer freeing slabs to RCU */
Paul Jackson101a5002006-03-24 03:16:07 -080066#define SLAB_MEM_SPREAD 0x00100000UL /* Spread some memory over cpuset */
Christoph Lameter81819f02007-05-06 14:49:36 -070067#define SLAB_TRACE 0x00200000UL /* Trace allocations and frees */
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
Thomas Gleixner30327ac2008-04-30 00:54:59 -070069/* Flag to prevent checks on free */
70#ifdef CONFIG_DEBUG_OBJECTS
71# define SLAB_DEBUG_OBJECTS 0x00400000UL
72#else
73# define SLAB_DEBUG_OBJECTS 0x00000000UL
74#endif
75
Catalin Marinasd5cff632009-06-11 13:22:40 +010076#define SLAB_NOLEAKTRACE 0x00800000UL /* Avoid kmemleak tracing */
77
Vegard Nossum2dff4402008-05-31 15:56:17 +020078/* Don't track use of uninitialized memory */
79#ifdef CONFIG_KMEMCHECK
80# define SLAB_NOTRACK 0x01000000UL
81#else
82# define SLAB_NOTRACK 0x00000000UL
83#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +030084#ifdef CONFIG_FAILSLAB
85# define SLAB_FAILSLAB 0x02000000UL /* Fault injection mark */
86#else
87# define SLAB_FAILSLAB 0x00000000UL
88#endif
Johannes Weiner127424c2016-01-20 15:02:32 -080089#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
Vladimir Davydov230e9fc2016-01-14 15:18:15 -080090# define SLAB_ACCOUNT 0x04000000UL /* Account to memcg */
91#else
92# define SLAB_ACCOUNT 0x00000000UL
93#endif
Vegard Nossum2dff4402008-05-31 15:56:17 +020094
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -070095#ifdef CONFIG_KASAN
96#define SLAB_KASAN 0x08000000UL
97#else
98#define SLAB_KASAN 0x00000000UL
99#endif
100
Mel Gormane12ba742007-10-16 01:25:52 -0700101/* The following flags affect the page allocator grouping pages by mobility */
102#define SLAB_RECLAIM_ACCOUNT 0x00020000UL /* Objects are reclaimable */
103#define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
Christoph Lameter2e892f42006-12-13 00:34:23 -0800104/*
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700105 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
106 *
107 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
108 *
109 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
110 * Both make kfree a no-op.
111 */
112#define ZERO_SIZE_PTR ((void *)16)
113
Roland Dreier1d4ec7b2007-07-20 12:13:20 -0700114#define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
Christoph Lameter6cb8f912007-07-17 04:03:22 -0700115 (unsigned long)ZERO_SIZE_PTR)
116
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000117#include <linux/kmemleak.h>
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800118#include <linux/kasan.h>
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500119
Glauber Costa2633d7a2012-12-18 14:22:34 -0800120struct mem_cgroup;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500121/*
Christoph Lameter2e892f42006-12-13 00:34:23 -0800122 * struct kmem_cache related prototypes
123 */
124void __init kmem_cache_init(void);
Denis Kirjanovfda90122015-11-05 18:44:59 -0800125bool slab_is_available(void);
Matt Mackall10cef602006-01-08 01:01:45 -0800126
Christoph Lameter2e892f42006-12-13 00:34:23 -0800127struct kmem_cache *kmem_cache_create(const char *, size_t, size_t,
Christoph Lameterebe29732006-12-06 20:32:59 -0800128 unsigned long,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700129 void (*)(void *));
Christoph Lameter2e892f42006-12-13 00:34:23 -0800130void kmem_cache_destroy(struct kmem_cache *);
131int kmem_cache_shrink(struct kmem_cache *);
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800132
133void memcg_create_kmem_cache(struct mem_cgroup *, struct kmem_cache *);
134void memcg_deactivate_kmem_caches(struct mem_cgroup *);
135void memcg_destroy_kmem_caches(struct mem_cgroup *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700137/*
138 * Please use this macro to create slab caches. Simply specify the
139 * name of the structure and maybe some flags that are listed above.
140 *
141 * The alignment of the struct determines object alignment. If you
142 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
143 * then the objects will be properly aligned in SMP configurations.
144 */
145#define KMEM_CACHE(__struct, __flags) kmem_cache_create(#__struct,\
146 sizeof(struct __struct), __alignof__(struct __struct),\
Paul Mundt20c2df82007-07-20 10:11:58 +0900147 (__flags), NULL)
Christoph Lameter0a31bd52007-05-06 14:49:57 -0700148
Christoph Lameter2e892f42006-12-13 00:34:23 -0800149/*
Christoph Lameter34504662013-01-10 19:00:53 +0000150 * Common kmalloc functions provided by all allocators
151 */
152void * __must_check __krealloc(const void *, size_t, gfp_t);
153void * __must_check krealloc(const void *, size_t, gfp_t);
154void kfree(const void *);
155void kzfree(const void *);
156size_t ksize(const void *);
157
Kees Cookf5509cc2016-06-07 11:05:33 -0700158#ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR
159const char *__check_heap_object(const void *ptr, unsigned long n,
160 struct page *page);
161#else
162static inline const char *__check_heap_object(const void *ptr,
163 unsigned long n,
164 struct page *page)
165{
166 return NULL;
167}
168#endif
169
Christoph Lameterc601fd62013-02-05 16:36:47 +0000170/*
171 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
172 * alignment larger than the alignment of a 64-bit integer.
173 * Setting ARCH_KMALLOC_MINALIGN in arch headers allows that.
174 */
175#if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
176#define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
177#define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
178#define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
179#else
180#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
181#endif
182
Christoph Lameter34504662013-01-10 19:00:53 +0000183/*
Rasmus Villemoes94a58c32015-11-20 15:56:48 -0800184 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
185 * Intended for arches that get misalignment faults even for 64 bit integer
186 * aligned buffers.
187 */
188#ifndef ARCH_SLAB_MINALIGN
189#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
190#endif
191
192/*
193 * kmalloc and friends return ARCH_KMALLOC_MINALIGN aligned
194 * pointers. kmem_cache_alloc and friends return ARCH_SLAB_MINALIGN
195 * aligned pointers.
196 */
197#define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
198#define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
199#define __assume_page_alignment __assume_aligned(PAGE_SIZE)
200
201/*
Christoph Lameter95a05b42013-01-10 19:14:19 +0000202 * Kmalloc array related definitions
203 */
204
205#ifdef CONFIG_SLAB
206/*
207 * The largest kmalloc size supported by the SLAB allocators is
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700208 * 32 megabyte (2^25) or the maximum allocatable page order if that is
209 * less than 32 MB.
210 *
211 * WARNING: Its not easy to increase this value since the allocators have
212 * to do various tricks to work around compiler limitations in order to
213 * ensure proper constant folding.
214 */
Christoph Lameterdebee072007-06-23 17:16:43 -0700215#define KMALLOC_SHIFT_HIGH ((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
216 (MAX_ORDER + PAGE_SHIFT - 1) : 25)
Christoph Lameter95a05b42013-01-10 19:14:19 +0000217#define KMALLOC_SHIFT_MAX KMALLOC_SHIFT_HIGH
Christoph Lameterc601fd62013-02-05 16:36:47 +0000218#ifndef KMALLOC_SHIFT_LOW
Christoph Lameter95a05b42013-01-10 19:14:19 +0000219#define KMALLOC_SHIFT_LOW 5
Christoph Lameterc601fd62013-02-05 16:36:47 +0000220#endif
Christoph Lameter069e2b352013-06-14 19:55:13 +0000221#endif
222
223#ifdef CONFIG_SLUB
Christoph Lameter95a05b42013-01-10 19:14:19 +0000224/*
Dave Hansen433a91f2014-01-28 14:24:50 -0800225 * SLUB directly allocates requests fitting in to an order-1 page
226 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
Christoph Lameter95a05b42013-01-10 19:14:19 +0000227 */
228#define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
229#define KMALLOC_SHIFT_MAX (MAX_ORDER + PAGE_SHIFT)
Christoph Lameterc601fd62013-02-05 16:36:47 +0000230#ifndef KMALLOC_SHIFT_LOW
Christoph Lameter95a05b42013-01-10 19:14:19 +0000231#define KMALLOC_SHIFT_LOW 3
232#endif
Christoph Lameterc601fd62013-02-05 16:36:47 +0000233#endif
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700234
Christoph Lameter069e2b352013-06-14 19:55:13 +0000235#ifdef CONFIG_SLOB
236/*
Dave Hansen433a91f2014-01-28 14:24:50 -0800237 * SLOB passes all requests larger than one page to the page allocator.
Christoph Lameter069e2b352013-06-14 19:55:13 +0000238 * No kmalloc array is necessary since objects of different sizes can
239 * be allocated from the same page.
240 */
Christoph Lameter069e2b352013-06-14 19:55:13 +0000241#define KMALLOC_SHIFT_HIGH PAGE_SHIFT
Dave Hansen433a91f2014-01-28 14:24:50 -0800242#define KMALLOC_SHIFT_MAX 30
Christoph Lameter069e2b352013-06-14 19:55:13 +0000243#ifndef KMALLOC_SHIFT_LOW
244#define KMALLOC_SHIFT_LOW 3
245#endif
246#endif
247
Christoph Lameter95a05b42013-01-10 19:14:19 +0000248/* Maximum allocatable size */
249#define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
250/* Maximum size for which we actually use a slab cache */
251#define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
252/* Maximum order allocatable via the slab allocagtor */
253#define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700254
Christoph Lameter90810642011-06-23 09:36:12 -0500255/*
Christoph Lameterce6a5022013-01-10 19:14:19 +0000256 * Kmalloc subsystem.
257 */
Christoph Lameterc601fd62013-02-05 16:36:47 +0000258#ifndef KMALLOC_MIN_SIZE
Christoph Lameter95a05b42013-01-10 19:14:19 +0000259#define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
Christoph Lameterce6a5022013-01-10 19:14:19 +0000260#endif
Christoph Lameterce6a5022013-01-10 19:14:19 +0000261
Joonsoo Kim24f870d2014-03-12 17:06:19 +0900262/*
263 * This restriction comes from byte sized index implementation.
264 * Page size is normally 2^12 bytes and, in this case, if we want to use
265 * byte sized index which can represent 2^8 entries, the size of the object
266 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
267 * If minimum size of kmalloc is less than 16, we use it as minimum object
268 * size and give up to use byte sized index.
269 */
270#define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
271 (KMALLOC_MIN_SIZE) : 16)
272
Christoph Lameter069e2b352013-06-14 19:55:13 +0000273#ifndef CONFIG_SLOB
Christoph Lameter9425c582013-01-10 19:12:17 +0000274extern struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
275#ifdef CONFIG_ZONE_DMA
276extern struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
277#endif
278
Christoph Lameterce6a5022013-01-10 19:14:19 +0000279/*
280 * Figure out which kmalloc slab an allocation of a certain size
281 * belongs to.
282 * 0 = zero alloc
283 * 1 = 65 .. 96 bytes
Rasmus Villemoes1ed58b62015-06-24 16:55:59 -0700284 * 2 = 129 .. 192 bytes
285 * n = 2^(n-1)+1 .. 2^n
Christoph Lameterce6a5022013-01-10 19:14:19 +0000286 */
287static __always_inline int kmalloc_index(size_t size)
288{
289 if (!size)
290 return 0;
291
292 if (size <= KMALLOC_MIN_SIZE)
293 return KMALLOC_SHIFT_LOW;
294
295 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
296 return 1;
297 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
298 return 2;
299 if (size <= 8) return 3;
300 if (size <= 16) return 4;
301 if (size <= 32) return 5;
302 if (size <= 64) return 6;
303 if (size <= 128) return 7;
304 if (size <= 256) return 8;
305 if (size <= 512) return 9;
306 if (size <= 1024) return 10;
307 if (size <= 2 * 1024) return 11;
308 if (size <= 4 * 1024) return 12;
309 if (size <= 8 * 1024) return 13;
310 if (size <= 16 * 1024) return 14;
311 if (size <= 32 * 1024) return 15;
312 if (size <= 64 * 1024) return 16;
313 if (size <= 128 * 1024) return 17;
314 if (size <= 256 * 1024) return 18;
315 if (size <= 512 * 1024) return 19;
316 if (size <= 1024 * 1024) return 20;
317 if (size <= 2 * 1024 * 1024) return 21;
318 if (size <= 4 * 1024 * 1024) return 22;
319 if (size <= 8 * 1024 * 1024) return 23;
320 if (size <= 16 * 1024 * 1024) return 24;
321 if (size <= 32 * 1024 * 1024) return 25;
322 if (size <= 64 * 1024 * 1024) return 26;
323 BUG();
324
325 /* Will never be reached. Needed because the compiler may complain */
326 return -1;
327}
Christoph Lameter069e2b352013-06-14 19:55:13 +0000328#endif /* !CONFIG_SLOB */
Christoph Lameterce6a5022013-01-10 19:14:19 +0000329
Rasmus Villemoes48a27052016-05-19 17:10:55 -0700330void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __malloc;
331void *kmem_cache_alloc(struct kmem_cache *, gfp_t flags) __assume_slab_alignment __malloc;
Vladimir Davydov2a4db7e2015-02-12 14:59:32 -0800332void kmem_cache_free(struct kmem_cache *, void *);
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000333
Christoph Lameter484748f2015-09-04 15:45:34 -0700334/*
Jesper Dangaard Brouer9f706d62016-03-15 14:54:03 -0700335 * Bulk allocation and freeing operations. These are accelerated in an
Christoph Lameter484748f2015-09-04 15:45:34 -0700336 * allocator specific way to avoid taking locks repeatedly or building
337 * metadata structures unnecessarily.
338 *
339 * Note that interrupts must be enabled when calling these functions.
340 */
341void kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -0800342int kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
Christoph Lameter484748f2015-09-04 15:45:34 -0700343
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -0700344/*
345 * Caller must not use kfree_bulk() on memory not originally allocated
346 * by kmalloc(), because the SLOB allocator cannot handle this.
347 */
348static __always_inline void kfree_bulk(size_t size, void **p)
349{
350 kmem_cache_free_bulk(NULL, size, p);
351}
352
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000353#ifdef CONFIG_NUMA
Rasmus Villemoes48a27052016-05-19 17:10:55 -0700354void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment __malloc;
355void *kmem_cache_alloc_node(struct kmem_cache *, gfp_t flags, int node) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000356#else
357static __always_inline void *__kmalloc_node(size_t size, gfp_t flags, int node)
358{
359 return __kmalloc(size, flags);
360}
361
362static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
363{
364 return kmem_cache_alloc(s, flags);
365}
366#endif
367
368#ifdef CONFIG_TRACING
Rasmus Villemoes48a27052016-05-19 17:10:55 -0700369extern void *kmem_cache_alloc_trace(struct kmem_cache *, gfp_t, size_t) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000370
371#ifdef CONFIG_NUMA
372extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
373 gfp_t gfpflags,
Rasmus Villemoes48a27052016-05-19 17:10:55 -0700374 int node, size_t size) __assume_slab_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000375#else
376static __always_inline void *
377kmem_cache_alloc_node_trace(struct kmem_cache *s,
378 gfp_t gfpflags,
379 int node, size_t size)
380{
381 return kmem_cache_alloc_trace(s, gfpflags, size);
382}
383#endif /* CONFIG_NUMA */
384
385#else /* CONFIG_TRACING */
386static __always_inline void *kmem_cache_alloc_trace(struct kmem_cache *s,
387 gfp_t flags, size_t size)
388{
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800389 void *ret = kmem_cache_alloc(s, flags);
390
Alexander Potapenko505f5dc2016-03-25 14:22:02 -0700391 kasan_kmalloc(s, ret, size, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800392 return ret;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000393}
394
395static __always_inline void *
396kmem_cache_alloc_node_trace(struct kmem_cache *s,
397 gfp_t gfpflags,
398 int node, size_t size)
399{
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800400 void *ret = kmem_cache_alloc_node(s, gfpflags, node);
401
Alexander Potapenko505f5dc2016-03-25 14:22:02 -0700402 kasan_kmalloc(s, ret, size, gfpflags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -0800403 return ret;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000404}
405#endif /* CONFIG_TRACING */
406
Rasmus Villemoes48a27052016-05-19 17:10:55 -0700407extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000408
409#ifdef CONFIG_TRACING
Rasmus Villemoes48a27052016-05-19 17:10:55 -0700410extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment __malloc;
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000411#else
412static __always_inline void *
413kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
414{
415 return kmalloc_order(size, flags, order);
416}
Christoph Lameterce6a5022013-01-10 19:14:19 +0000417#endif
418
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000419static __always_inline void *kmalloc_large(size_t size, gfp_t flags)
420{
421 unsigned int order = get_order(size);
422 return kmalloc_order_trace(size, flags, order);
423}
424
425/**
426 * kmalloc - allocate memory
427 * @size: how many bytes of memory are required.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800428 * @flags: the type of memory to allocate.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000429 *
430 * kmalloc is the normal method of allocating memory
431 * for objects smaller than page size in the kernel.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800432 *
433 * The @flags argument may be one of:
434 *
435 * %GFP_USER - Allocate memory on behalf of user. May sleep.
436 *
437 * %GFP_KERNEL - Allocate normal kernel ram. May sleep.
438 *
439 * %GFP_ATOMIC - Allocation will not sleep. May use emergency pools.
440 * For example, use this inside interrupt handlers.
441 *
442 * %GFP_HIGHUSER - Allocate pages from high memory.
443 *
444 * %GFP_NOIO - Do not do any I/O at all while trying to get memory.
445 *
446 * %GFP_NOFS - Do not make any fs calls while trying to get memory.
447 *
448 * %GFP_NOWAIT - Allocation will not sleep.
449 *
Johannes Weinere97ca8e2014-03-10 15:49:43 -0700450 * %__GFP_THISNODE - Allocate node-local memory only.
Randy Dunlap7e3528c2013-11-22 18:14:38 -0800451 *
452 * %GFP_DMA - Allocation suitable for DMA.
453 * Should only be used for kmalloc() caches. Otherwise, use a
454 * slab created with SLAB_DMA.
455 *
456 * Also it is possible to set different flags by OR'ing
457 * in one or more of the following additional @flags:
458 *
459 * %__GFP_COLD - Request cache-cold pages instead of
460 * trying to return cache-warm pages.
461 *
462 * %__GFP_HIGH - This allocation has high priority and may use emergency pools.
463 *
464 * %__GFP_NOFAIL - Indicate that this allocation is in no way allowed to fail
465 * (think twice before using).
466 *
467 * %__GFP_NORETRY - If memory is not immediately available,
468 * then give up at once.
469 *
470 * %__GFP_NOWARN - If allocation fails, don't issue any warnings.
471 *
472 * %__GFP_REPEAT - If allocation fails initially, try once more before failing.
473 *
474 * There are other flags available as well, but these are not intended
475 * for general use, and so are not documented here. For a full list of
476 * potential flags, always refer to linux/gfp.h.
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000477 */
478static __always_inline void *kmalloc(size_t size, gfp_t flags)
479{
480 if (__builtin_constant_p(size)) {
481 if (size > KMALLOC_MAX_CACHE_SIZE)
482 return kmalloc_large(size, flags);
483#ifndef CONFIG_SLOB
484 if (!(flags & GFP_DMA)) {
485 int index = kmalloc_index(size);
486
487 if (!index)
488 return ZERO_SIZE_PTR;
489
490 return kmem_cache_alloc_trace(kmalloc_caches[index],
491 flags, size);
492 }
493#endif
494 }
495 return __kmalloc(size, flags);
496}
497
Christoph Lameterce6a5022013-01-10 19:14:19 +0000498/*
499 * Determine size used for the nth kmalloc cache.
500 * return size or 0 if a kmalloc cache for that
501 * size does not exist
502 */
503static __always_inline int kmalloc_size(int n)
504{
Christoph Lameter069e2b352013-06-14 19:55:13 +0000505#ifndef CONFIG_SLOB
Christoph Lameterce6a5022013-01-10 19:14:19 +0000506 if (n > 2)
507 return 1 << n;
508
509 if (n == 1 && KMALLOC_MIN_SIZE <= 32)
510 return 96;
511
512 if (n == 2 && KMALLOC_MIN_SIZE <= 64)
513 return 192;
Christoph Lameter069e2b352013-06-14 19:55:13 +0000514#endif
Christoph Lameterce6a5022013-01-10 19:14:19 +0000515 return 0;
516}
Christoph Lameterce6a5022013-01-10 19:14:19 +0000517
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000518static __always_inline void *kmalloc_node(size_t size, gfp_t flags, int node)
519{
520#ifndef CONFIG_SLOB
521 if (__builtin_constant_p(size) &&
Christoph Lameter23774a22013-09-04 19:58:08 +0000522 size <= KMALLOC_MAX_CACHE_SIZE && !(flags & GFP_DMA)) {
Christoph Lameterf1b6eb62013-09-04 16:35:34 +0000523 int i = kmalloc_index(size);
524
525 if (!i)
526 return ZERO_SIZE_PTR;
527
528 return kmem_cache_alloc_node_trace(kmalloc_caches[i],
529 flags, node, size);
530 }
531#endif
532 return __kmalloc_node(size, flags, node);
533}
534
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800535struct memcg_cache_array {
536 struct rcu_head rcu;
537 struct kmem_cache *entries[0];
538};
539
Christoph Lameter0aa817f2007-05-16 22:11:01 -0700540/*
Glauber Costaba6c4962012-12-18 14:22:27 -0800541 * This is the main placeholder for memcg-related information in kmem caches.
Glauber Costaba6c4962012-12-18 14:22:27 -0800542 * Both the root cache and the child caches will have it. For the root cache,
543 * this will hold a dynamically allocated array large enough to hold
Vladimir Davydovf8570262014-01-23 15:53:06 -0800544 * information about the currently limited memcgs in the system. To allow the
545 * array to be accessed without taking any locks, on relocation we free the old
546 * version only after a grace period.
Glauber Costaba6c4962012-12-18 14:22:27 -0800547 *
548 * Child caches will hold extra metadata needed for its operation. Fields are:
549 *
550 * @memcg: pointer to the memcg this cache belongs to
Glauber Costa2633d7a2012-12-18 14:22:34 -0800551 * @root_cache: pointer to the global, root cache, this cache was derived from
Vladimir Davydov426589f2015-02-12 14:59:23 -0800552 *
553 * Both root and child caches of the same kind are linked into a list chained
554 * through @list.
Glauber Costaba6c4962012-12-18 14:22:27 -0800555 */
556struct memcg_cache_params {
557 bool is_root_cache;
Vladimir Davydov426589f2015-02-12 14:59:23 -0800558 struct list_head list;
Glauber Costaba6c4962012-12-18 14:22:27 -0800559 union {
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800560 struct memcg_cache_array __rcu *memcg_caches;
Glauber Costa2633d7a2012-12-18 14:22:34 -0800561 struct {
562 struct mem_cgroup *memcg;
Glauber Costa2633d7a2012-12-18 14:22:34 -0800563 struct kmem_cache *root_cache;
564 };
Glauber Costaba6c4962012-12-18 14:22:27 -0800565 };
566};
567
Glauber Costa2633d7a2012-12-18 14:22:34 -0800568int memcg_update_all_caches(int num_memcgs);
569
Christoph Lameter2e892f42006-12-13 00:34:23 -0800570/**
Michael Opdenackere7efa612013-06-25 18:16:55 +0200571 * kmalloc_array - allocate memory for an array.
572 * @n: number of elements.
573 * @size: element size.
574 * @flags: the type of memory to allocate (see kmalloc).
Paul Drynoff800590f2006-06-23 02:03:48 -0700575 */
Xi Wanga8203722012-03-05 15:14:41 -0800576static inline void *kmalloc_array(size_t n, size_t size, gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577{
Xi Wanga3860c12012-05-31 16:26:04 -0700578 if (size != 0 && n > SIZE_MAX / size)
Paul Mundt6193a2f2007-07-15 23:38:22 -0700579 return NULL;
Alexey Dobriyan91c6a052016-07-26 15:22:08 -0700580 if (__builtin_constant_p(n) && __builtin_constant_p(size))
581 return kmalloc(n * size, flags);
Xi Wanga8203722012-03-05 15:14:41 -0800582 return __kmalloc(n * size, flags);
583}
584
585/**
586 * kcalloc - allocate memory for an array. The memory is set to zero.
587 * @n: number of elements.
588 * @size: element size.
589 * @flags: the type of memory to allocate (see kmalloc).
590 */
591static inline void *kcalloc(size_t n, size_t size, gfp_t flags)
592{
593 return kmalloc_array(n, size, flags | __GFP_ZERO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594}
595
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700596/*
597 * kmalloc_track_caller is a special version of kmalloc that records the
598 * calling function of the routine calling it for slab leak tracking instead
599 * of just the calling function (confusing, eh?).
600 * It's useful when the call to kmalloc comes from a widely-used standard
601 * allocator where we care about the real place the memory allocation
602 * request comes from.
603 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300604extern void *__kmalloc_track_caller(size_t, gfp_t, unsigned long);
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -0700605#define kmalloc_track_caller(size, flags) \
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300606 __kmalloc_track_caller(size, flags, _RET_IP_)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
Manfred Spraul97e2bde2005-05-01 08:58:38 -0700608#ifdef CONFIG_NUMA
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300609extern void *__kmalloc_node_track_caller(size_t, gfp_t, int, unsigned long);
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800610#define kmalloc_node_track_caller(size, flags, node) \
611 __kmalloc_node_track_caller(size, flags, node, \
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300612 _RET_IP_)
Christoph Lameter2e892f42006-12-13 00:34:23 -0800613
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800614#else /* CONFIG_NUMA */
Christoph Lameter2e892f42006-12-13 00:34:23 -0800615
616#define kmalloc_node_track_caller(size, flags, node) \
617 kmalloc_track_caller(size, flags)
618
Pascal Terjandfcd3612008-11-25 15:08:19 +0100619#endif /* CONFIG_NUMA */
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800620
Christoph Lameter81cda662007-07-17 04:03:29 -0700621/*
622 * Shortcuts
623 */
624static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
625{
626 return kmem_cache_alloc(k, flags | __GFP_ZERO);
627}
628
629/**
630 * kzalloc - allocate memory. The memory is set to zero.
631 * @size: how many bytes of memory are required.
632 * @flags: the type of memory to allocate (see kmalloc).
633 */
634static inline void *kzalloc(size_t size, gfp_t flags)
635{
636 return kmalloc(size, flags | __GFP_ZERO);
637}
638
Jeff Layton979b0fe2008-06-05 22:47:00 -0700639/**
640 * kzalloc_node - allocate zeroed memory from a particular memory node.
641 * @size: how many bytes of memory are required.
642 * @flags: the type of memory to allocate (see kmalloc).
643 * @node: memory node from which to allocate
644 */
645static inline void *kzalloc_node(size_t size, gfp_t flags, int node)
646{
647 return kmalloc_node(size, flags | __GFP_ZERO, node);
648}
649
Joonsoo Kim07f361b2014-10-09 15:26:00 -0700650unsigned int kmem_cache_size(struct kmem_cache *s);
Pekka Enberg7e85ee02009-06-12 14:03:06 +0300651void __init kmem_cache_init_late(void);
652
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +0200653#if defined(CONFIG_SMP) && defined(CONFIG_SLAB)
654int slab_prepare_cpu(unsigned int cpu);
655int slab_dead_cpu(unsigned int cpu);
656#else
657#define slab_prepare_cpu NULL
658#define slab_dead_cpu NULL
659#endif
660
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661#endif /* _LINUX_SLAB_H */