xref: /linux-6.15/include/linux/slab.h (revision fc5dfebc)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3  * Written by Mark Hemment, 1996 ([email protected]).
4  *
5  * (C) SGI 2006, Christoph Lameter
6  * 	Cleaned up and restructured to ease the addition of alternative
7  * 	implementations of SLAB allocators.
8  * (C) Linux Foundation 2008-2013
9  *      Unified interface for all slab allocators
10  */
11 
12 #ifndef _LINUX_SLAB_H
13 #define	_LINUX_SLAB_H
14 
15 #include <linux/gfp.h>
16 #include <linux/overflow.h>
17 #include <linux/types.h>
18 #include <linux/workqueue.h>
19 #include <linux/percpu-refcount.h>
20 
21 
22 /*
23  * Flags to pass to kmem_cache_create().
24  * The ones marked DEBUG are only valid if CONFIG_DEBUG_SLAB is set.
25  */
26 /* DEBUG: Perform (expensive) checks on alloc/free */
27 #define SLAB_CONSISTENCY_CHECKS	((slab_flags_t __force)0x00000100U)
28 /* DEBUG: Red zone objs in a cache */
29 #define SLAB_RED_ZONE		((slab_flags_t __force)0x00000400U)
30 /* DEBUG: Poison objects */
31 #define SLAB_POISON		((slab_flags_t __force)0x00000800U)
32 /* Align objs on cache lines */
33 #define SLAB_HWCACHE_ALIGN	((slab_flags_t __force)0x00002000U)
34 /* Use GFP_DMA memory */
35 #define SLAB_CACHE_DMA		((slab_flags_t __force)0x00004000U)
36 /* Use GFP_DMA32 memory */
37 #define SLAB_CACHE_DMA32	((slab_flags_t __force)0x00008000U)
38 /* DEBUG: Store the last owner for bug hunting */
39 #define SLAB_STORE_USER		((slab_flags_t __force)0x00010000U)
40 /* Panic if kmem_cache_create() fails */
41 #define SLAB_PANIC		((slab_flags_t __force)0x00040000U)
42 /*
43  * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS!
44  *
45  * This delays freeing the SLAB page by a grace period, it does _NOT_
46  * delay object freeing. This means that if you do kmem_cache_free()
47  * that memory location is free to be reused at any time. Thus it may
48  * be possible to see another object there in the same RCU grace period.
49  *
50  * This feature only ensures the memory location backing the object
51  * stays valid, the trick to using this is relying on an independent
52  * object validation pass. Something like:
53  *
54  *  rcu_read_lock()
55  * again:
56  *  obj = lockless_lookup(key);
57  *  if (obj) {
58  *    if (!try_get_ref(obj)) // might fail for free objects
59  *      goto again;
60  *
61  *    if (obj->key != key) { // not the object we expected
62  *      put_ref(obj);
63  *      goto again;
64  *    }
65  *  }
66  *  rcu_read_unlock();
67  *
68  * This is useful if we need to approach a kernel structure obliquely,
69  * from its address obtained without the usual locking. We can lock
70  * the structure to stabilize it and check it's still at the given address,
71  * only if we can be sure that the memory has not been meanwhile reused
72  * for some other kind of object (which our subsystem's lock might corrupt).
73  *
74  * rcu_read_lock before reading the address, then rcu_read_unlock after
75  * taking the spinlock within the structure expected at that address.
76  *
77  * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU.
78  */
79 /* Defer freeing slabs to RCU */
80 #define SLAB_TYPESAFE_BY_RCU	((slab_flags_t __force)0x00080000U)
81 /* Spread some memory over cpuset */
82 #define SLAB_MEM_SPREAD		((slab_flags_t __force)0x00100000U)
83 /* Trace allocations and frees */
84 #define SLAB_TRACE		((slab_flags_t __force)0x00200000U)
85 
86 /* Flag to prevent checks on free */
87 #ifdef CONFIG_DEBUG_OBJECTS
88 # define SLAB_DEBUG_OBJECTS	((slab_flags_t __force)0x00400000U)
89 #else
90 # define SLAB_DEBUG_OBJECTS	0
91 #endif
92 
93 /* Avoid kmemleak tracing */
94 #define SLAB_NOLEAKTRACE	((slab_flags_t __force)0x00800000U)
95 
96 /* Fault injection mark */
97 #ifdef CONFIG_FAILSLAB
98 # define SLAB_FAILSLAB		((slab_flags_t __force)0x02000000U)
99 #else
100 # define SLAB_FAILSLAB		0
101 #endif
102 /* Account to memcg */
103 #ifdef CONFIG_MEMCG_KMEM
104 # define SLAB_ACCOUNT		((slab_flags_t __force)0x04000000U)
105 #else
106 # define SLAB_ACCOUNT		0
107 #endif
108 
109 #ifdef CONFIG_KASAN_GENERIC
110 #define SLAB_KASAN		((slab_flags_t __force)0x08000000U)
111 #else
112 #define SLAB_KASAN		0
113 #endif
114 
115 /*
116  * Ignore user specified debugging flags.
117  * Intended for caches created for self-tests so they have only flags
118  * specified in the code and other flags are ignored.
119  */
120 #define SLAB_NO_USER_FLAGS	((slab_flags_t __force)0x10000000U)
121 
122 #ifdef CONFIG_KFENCE
123 #define SLAB_SKIP_KFENCE	((slab_flags_t __force)0x20000000U)
124 #else
125 #define SLAB_SKIP_KFENCE	0
126 #endif
127 
128 /* The following flags affect the page allocator grouping pages by mobility */
129 /* Objects are reclaimable */
130 #define SLAB_RECLAIM_ACCOUNT	((slab_flags_t __force)0x00020000U)
131 #define SLAB_TEMPORARY		SLAB_RECLAIM_ACCOUNT	/* Objects are short-lived */
132 
133 /*
134  * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
135  *
136  * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
137  *
138  * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
139  * Both make kfree a no-op.
140  */
141 #define ZERO_SIZE_PTR ((void *)16)
142 
143 #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
144 				(unsigned long)ZERO_SIZE_PTR)
145 
146 #include <linux/kasan.h>
147 
148 struct list_lru;
149 struct mem_cgroup;
150 /*
151  * struct kmem_cache related prototypes
152  */
153 void __init kmem_cache_init(void);
154 bool slab_is_available(void);
155 
156 struct kmem_cache *kmem_cache_create(const char *name, unsigned int size,
157 			unsigned int align, slab_flags_t flags,
158 			void (*ctor)(void *));
159 struct kmem_cache *kmem_cache_create_usercopy(const char *name,
160 			unsigned int size, unsigned int align,
161 			slab_flags_t flags,
162 			unsigned int useroffset, unsigned int usersize,
163 			void (*ctor)(void *));
164 void kmem_cache_destroy(struct kmem_cache *s);
165 int kmem_cache_shrink(struct kmem_cache *s);
166 
167 /*
168  * Please use this macro to create slab caches. Simply specify the
169  * name of the structure and maybe some flags that are listed above.
170  *
171  * The alignment of the struct determines object alignment. If you
172  * f.e. add ____cacheline_aligned_in_smp to the struct declaration
173  * then the objects will be properly aligned in SMP configurations.
174  */
175 #define KMEM_CACHE(__struct, __flags)					\
176 		kmem_cache_create(#__struct, sizeof(struct __struct),	\
177 			__alignof__(struct __struct), (__flags), NULL)
178 
179 /*
180  * To whitelist a single field for copying to/from usercopy, use this
181  * macro instead for KMEM_CACHE() above.
182  */
183 #define KMEM_CACHE_USERCOPY(__struct, __flags, __field)			\
184 		kmem_cache_create_usercopy(#__struct,			\
185 			sizeof(struct __struct),			\
186 			__alignof__(struct __struct), (__flags),	\
187 			offsetof(struct __struct, __field),		\
188 			sizeof_field(struct __struct, __field), NULL)
189 
190 /*
191  * Common kmalloc functions provided by all allocators
192  */
193 void * __must_check krealloc(const void *objp, size_t new_size, gfp_t flags) __alloc_size(2);
194 void kfree(const void *objp);
195 void kfree_sensitive(const void *objp);
196 size_t __ksize(const void *objp);
197 size_t ksize(const void *objp);
198 #ifdef CONFIG_PRINTK
199 bool kmem_valid_obj(void *object);
200 void kmem_dump_obj(void *object);
201 #endif
202 
203 /*
204  * Some archs want to perform DMA into kmalloc caches and need a guaranteed
205  * alignment larger than the alignment of a 64-bit integer.
206  * Setting ARCH_DMA_MINALIGN in arch headers allows that.
207  */
208 #if defined(ARCH_DMA_MINALIGN) && ARCH_DMA_MINALIGN > 8
209 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
210 #define KMALLOC_MIN_SIZE ARCH_DMA_MINALIGN
211 #define KMALLOC_SHIFT_LOW ilog2(ARCH_DMA_MINALIGN)
212 #else
213 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
214 #endif
215 
216 /*
217  * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
218  * Intended for arches that get misalignment faults even for 64 bit integer
219  * aligned buffers.
220  */
221 #ifndef ARCH_SLAB_MINALIGN
222 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
223 #endif
224 
225 /*
226  * Arches can define this function if they want to decide the minimum slab
227  * alignment at runtime. The value returned by the function must be a power
228  * of two and >= ARCH_SLAB_MINALIGN.
229  */
230 #ifndef arch_slab_minalign
231 static inline unsigned int arch_slab_minalign(void)
232 {
233 	return ARCH_SLAB_MINALIGN;
234 }
235 #endif
236 
237 /*
238  * kmem_cache_alloc and friends return pointers aligned to ARCH_SLAB_MINALIGN.
239  * kmalloc and friends return pointers aligned to both ARCH_KMALLOC_MINALIGN
240  * and ARCH_SLAB_MINALIGN, but here we only assume the former alignment.
241  */
242 #define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
243 #define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
244 #define __assume_page_alignment __assume_aligned(PAGE_SIZE)
245 
246 /*
247  * Kmalloc array related definitions
248  */
249 
250 #ifdef CONFIG_SLAB
251 /*
252  * The largest kmalloc size supported by the SLAB allocators is
253  * 32 megabyte (2^25) or the maximum allocatable page order if that is
254  * less than 32 MB.
255  *
256  * WARNING: Its not easy to increase this value since the allocators have
257  * to do various tricks to work around compiler limitations in order to
258  * ensure proper constant folding.
259  */
260 #define KMALLOC_SHIFT_HIGH	((MAX_ORDER + PAGE_SHIFT - 1) <= 25 ? \
261 				(MAX_ORDER + PAGE_SHIFT - 1) : 25)
262 #define KMALLOC_SHIFT_MAX	KMALLOC_SHIFT_HIGH
263 #ifndef KMALLOC_SHIFT_LOW
264 #define KMALLOC_SHIFT_LOW	5
265 #endif
266 #endif
267 
268 #ifdef CONFIG_SLUB
269 /*
270  * SLUB directly allocates requests fitting in to an order-1 page
271  * (PAGE_SIZE*2).  Larger requests are passed to the page allocator.
272  */
273 #define KMALLOC_SHIFT_HIGH	(PAGE_SHIFT + 1)
274 #define KMALLOC_SHIFT_MAX	(MAX_ORDER + PAGE_SHIFT - 1)
275 #ifndef KMALLOC_SHIFT_LOW
276 #define KMALLOC_SHIFT_LOW	3
277 #endif
278 #endif
279 
280 #ifdef CONFIG_SLOB
281 /*
282  * SLOB passes all requests larger than one page to the page allocator.
283  * No kmalloc array is necessary since objects of different sizes can
284  * be allocated from the same page.
285  */
286 #define KMALLOC_SHIFT_HIGH	PAGE_SHIFT
287 #define KMALLOC_SHIFT_MAX	(MAX_ORDER + PAGE_SHIFT - 1)
288 #ifndef KMALLOC_SHIFT_LOW
289 #define KMALLOC_SHIFT_LOW	3
290 #endif
291 #endif
292 
293 /* Maximum allocatable size */
294 #define KMALLOC_MAX_SIZE	(1UL << KMALLOC_SHIFT_MAX)
295 /* Maximum size for which we actually use a slab cache */
296 #define KMALLOC_MAX_CACHE_SIZE	(1UL << KMALLOC_SHIFT_HIGH)
297 /* Maximum order allocatable via the slab allocator */
298 #define KMALLOC_MAX_ORDER	(KMALLOC_SHIFT_MAX - PAGE_SHIFT)
299 
300 /*
301  * Kmalloc subsystem.
302  */
303 #ifndef KMALLOC_MIN_SIZE
304 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
305 #endif
306 
307 /*
308  * This restriction comes from byte sized index implementation.
309  * Page size is normally 2^12 bytes and, in this case, if we want to use
310  * byte sized index which can represent 2^8 entries, the size of the object
311  * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
312  * If minimum size of kmalloc is less than 16, we use it as minimum object
313  * size and give up to use byte sized index.
314  */
315 #define SLAB_OBJ_MIN_SIZE      (KMALLOC_MIN_SIZE < 16 ? \
316                                (KMALLOC_MIN_SIZE) : 16)
317 
318 /*
319  * Whenever changing this, take care of that kmalloc_type() and
320  * create_kmalloc_caches() still work as intended.
321  *
322  * KMALLOC_NORMAL can contain only unaccounted objects whereas KMALLOC_CGROUP
323  * is for accounted but unreclaimable and non-dma objects. All the other
324  * kmem caches can have both accounted and unaccounted objects.
325  */
326 enum kmalloc_cache_type {
327 	KMALLOC_NORMAL = 0,
328 #ifndef CONFIG_ZONE_DMA
329 	KMALLOC_DMA = KMALLOC_NORMAL,
330 #endif
331 #ifndef CONFIG_MEMCG_KMEM
332 	KMALLOC_CGROUP = KMALLOC_NORMAL,
333 #else
334 	KMALLOC_CGROUP,
335 #endif
336 	KMALLOC_RECLAIM,
337 #ifdef CONFIG_ZONE_DMA
338 	KMALLOC_DMA,
339 #endif
340 	NR_KMALLOC_TYPES
341 };
342 
343 #ifndef CONFIG_SLOB
344 extern struct kmem_cache *
345 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1];
346 
347 /*
348  * Define gfp bits that should not be set for KMALLOC_NORMAL.
349  */
350 #define KMALLOC_NOT_NORMAL_BITS					\
351 	(__GFP_RECLAIMABLE |					\
352 	(IS_ENABLED(CONFIG_ZONE_DMA)   ? __GFP_DMA : 0) |	\
353 	(IS_ENABLED(CONFIG_MEMCG_KMEM) ? __GFP_ACCOUNT : 0))
354 
355 static __always_inline enum kmalloc_cache_type kmalloc_type(gfp_t flags)
356 {
357 	/*
358 	 * The most common case is KMALLOC_NORMAL, so test for it
359 	 * with a single branch for all the relevant flags.
360 	 */
361 	if (likely((flags & KMALLOC_NOT_NORMAL_BITS) == 0))
362 		return KMALLOC_NORMAL;
363 
364 	/*
365 	 * At least one of the flags has to be set. Their priorities in
366 	 * decreasing order are:
367 	 *  1) __GFP_DMA
368 	 *  2) __GFP_RECLAIMABLE
369 	 *  3) __GFP_ACCOUNT
370 	 */
371 	if (IS_ENABLED(CONFIG_ZONE_DMA) && (flags & __GFP_DMA))
372 		return KMALLOC_DMA;
373 	if (!IS_ENABLED(CONFIG_MEMCG_KMEM) || (flags & __GFP_RECLAIMABLE))
374 		return KMALLOC_RECLAIM;
375 	else
376 		return KMALLOC_CGROUP;
377 }
378 
379 /*
380  * Figure out which kmalloc slab an allocation of a certain size
381  * belongs to.
382  * 0 = zero alloc
383  * 1 =  65 .. 96 bytes
384  * 2 = 129 .. 192 bytes
385  * n = 2^(n-1)+1 .. 2^n
386  *
387  * Note: __kmalloc_index() is compile-time optimized, and not runtime optimized;
388  * typical usage is via kmalloc_index() and therefore evaluated at compile-time.
389  * Callers where !size_is_constant should only be test modules, where runtime
390  * overheads of __kmalloc_index() can be tolerated.  Also see kmalloc_slab().
391  */
392 static __always_inline unsigned int __kmalloc_index(size_t size,
393 						    bool size_is_constant)
394 {
395 	if (!size)
396 		return 0;
397 
398 	if (size <= KMALLOC_MIN_SIZE)
399 		return KMALLOC_SHIFT_LOW;
400 
401 	if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
402 		return 1;
403 	if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
404 		return 2;
405 	if (size <=          8) return 3;
406 	if (size <=         16) return 4;
407 	if (size <=         32) return 5;
408 	if (size <=         64) return 6;
409 	if (size <=        128) return 7;
410 	if (size <=        256) return 8;
411 	if (size <=        512) return 9;
412 	if (size <=       1024) return 10;
413 	if (size <=   2 * 1024) return 11;
414 	if (size <=   4 * 1024) return 12;
415 	if (size <=   8 * 1024) return 13;
416 	if (size <=  16 * 1024) return 14;
417 	if (size <=  32 * 1024) return 15;
418 	if (size <=  64 * 1024) return 16;
419 	if (size <= 128 * 1024) return 17;
420 	if (size <= 256 * 1024) return 18;
421 	if (size <= 512 * 1024) return 19;
422 	if (size <= 1024 * 1024) return 20;
423 	if (size <=  2 * 1024 * 1024) return 21;
424 	if (size <=  4 * 1024 * 1024) return 22;
425 	if (size <=  8 * 1024 * 1024) return 23;
426 	if (size <=  16 * 1024 * 1024) return 24;
427 	if (size <=  32 * 1024 * 1024) return 25;
428 
429 	if (!IS_ENABLED(CONFIG_PROFILE_ALL_BRANCHES) && size_is_constant)
430 		BUILD_BUG_ON_MSG(1, "unexpected size in kmalloc_index()");
431 	else
432 		BUG();
433 
434 	/* Will never be reached. Needed because the compiler may complain */
435 	return -1;
436 }
437 #define kmalloc_index(s) __kmalloc_index(s, true)
438 #endif /* !CONFIG_SLOB */
439 
440 void *__kmalloc(size_t size, gfp_t flags) __assume_kmalloc_alignment __alloc_size(1);
441 void *kmem_cache_alloc(struct kmem_cache *s, gfp_t flags) __assume_slab_alignment __malloc;
442 void *kmem_cache_alloc_lru(struct kmem_cache *s, struct list_lru *lru,
443 			   gfp_t gfpflags) __assume_slab_alignment __malloc;
444 void kmem_cache_free(struct kmem_cache *s, void *objp);
445 
446 /*
447  * Bulk allocation and freeing operations. These are accelerated in an
448  * allocator specific way to avoid taking locks repeatedly or building
449  * metadata structures unnecessarily.
450  *
451  * Note that interrupts must be enabled when calling these functions.
452  */
453 void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p);
454 int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, void **p);
455 
456 /*
457  * Caller must not use kfree_bulk() on memory not originally allocated
458  * by kmalloc(), because the SLOB allocator cannot handle this.
459  */
460 static __always_inline void kfree_bulk(size_t size, void **p)
461 {
462 	kmem_cache_free_bulk(NULL, size, p);
463 }
464 
465 #ifdef CONFIG_NUMA
466 void *__kmalloc_node(size_t size, gfp_t flags, int node) __assume_kmalloc_alignment
467 							 __alloc_size(1);
468 void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node) __assume_slab_alignment
469 									 __malloc;
470 #else
471 static __always_inline __alloc_size(1) void *__kmalloc_node(size_t size, gfp_t flags, int node)
472 {
473 	return __kmalloc(size, flags);
474 }
475 
476 static __always_inline void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t flags, int node)
477 {
478 	return kmem_cache_alloc(s, flags);
479 }
480 #endif
481 
482 #ifdef CONFIG_TRACING
483 extern void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t flags, size_t size)
484 				   __assume_slab_alignment __alloc_size(3);
485 
486 #ifdef CONFIG_NUMA
487 extern void *kmem_cache_alloc_node_trace(struct kmem_cache *s, gfp_t gfpflags,
488 					 int node, size_t size) __assume_slab_alignment
489 								__alloc_size(4);
490 #else
491 static __always_inline __alloc_size(4) void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
492 						 gfp_t gfpflags, int node, size_t size)
493 {
494 	return kmem_cache_alloc_trace(s, gfpflags, size);
495 }
496 #endif /* CONFIG_NUMA */
497 
498 #else /* CONFIG_TRACING */
499 static __always_inline __alloc_size(3) void *kmem_cache_alloc_trace(struct kmem_cache *s,
500 								    gfp_t flags, size_t size)
501 {
502 	void *ret = kmem_cache_alloc(s, flags);
503 
504 	ret = kasan_kmalloc(s, ret, size, flags);
505 	return ret;
506 }
507 
508 static __always_inline void *kmem_cache_alloc_node_trace(struct kmem_cache *s, gfp_t gfpflags,
509 							 int node, size_t size)
510 {
511 	void *ret = kmem_cache_alloc_node(s, gfpflags, node);
512 
513 	ret = kasan_kmalloc(s, ret, size, gfpflags);
514 	return ret;
515 }
516 #endif /* CONFIG_TRACING */
517 
518 extern void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) __assume_page_alignment
519 									 __alloc_size(1);
520 
521 #ifdef CONFIG_TRACING
522 extern void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
523 				__assume_page_alignment __alloc_size(1);
524 #else
525 static __always_inline __alloc_size(1) void *kmalloc_order_trace(size_t size, gfp_t flags,
526 								 unsigned int order)
527 {
528 	return kmalloc_order(size, flags, order);
529 }
530 #endif
531 
532 static __always_inline __alloc_size(1) void *kmalloc_large(size_t size, gfp_t flags)
533 {
534 	unsigned int order = get_order(size);
535 	return kmalloc_order_trace(size, flags, order);
536 }
537 
538 /**
539  * kmalloc - allocate memory
540  * @size: how many bytes of memory are required.
541  * @flags: the type of memory to allocate.
542  *
543  * kmalloc is the normal method of allocating memory
544  * for objects smaller than page size in the kernel.
545  *
546  * The allocated object address is aligned to at least ARCH_KMALLOC_MINALIGN
547  * bytes. For @size of power of two bytes, the alignment is also guaranteed
548  * to be at least to the size.
549  *
550  * The @flags argument may be one of the GFP flags defined at
551  * include/linux/gfp.h and described at
552  * :ref:`Documentation/core-api/mm-api.rst <mm-api-gfp-flags>`
553  *
554  * The recommended usage of the @flags is described at
555  * :ref:`Documentation/core-api/memory-allocation.rst <memory_allocation>`
556  *
557  * Below is a brief outline of the most useful GFP flags
558  *
559  * %GFP_KERNEL
560  *	Allocate normal kernel ram. May sleep.
561  *
562  * %GFP_NOWAIT
563  *	Allocation will not sleep.
564  *
565  * %GFP_ATOMIC
566  *	Allocation will not sleep.  May use emergency pools.
567  *
568  * %GFP_HIGHUSER
569  *	Allocate memory from high memory on behalf of user.
570  *
571  * Also it is possible to set different flags by OR'ing
572  * in one or more of the following additional @flags:
573  *
574  * %__GFP_HIGH
575  *	This allocation has high priority and may use emergency pools.
576  *
577  * %__GFP_NOFAIL
578  *	Indicate that this allocation is in no way allowed to fail
579  *	(think twice before using).
580  *
581  * %__GFP_NORETRY
582  *	If memory is not immediately available,
583  *	then give up at once.
584  *
585  * %__GFP_NOWARN
586  *	If allocation fails, don't issue any warnings.
587  *
588  * %__GFP_RETRY_MAYFAIL
589  *	Try really hard to succeed the allocation but fail
590  *	eventually.
591  */
592 static __always_inline __alloc_size(1) void *kmalloc(size_t size, gfp_t flags)
593 {
594 	if (__builtin_constant_p(size)) {
595 #ifndef CONFIG_SLOB
596 		unsigned int index;
597 #endif
598 		if (size > KMALLOC_MAX_CACHE_SIZE)
599 			return kmalloc_large(size, flags);
600 #ifndef CONFIG_SLOB
601 		index = kmalloc_index(size);
602 
603 		if (!index)
604 			return ZERO_SIZE_PTR;
605 
606 		return kmem_cache_alloc_trace(
607 				kmalloc_caches[kmalloc_type(flags)][index],
608 				flags, size);
609 #endif
610 	}
611 	return __kmalloc(size, flags);
612 }
613 
614 static __always_inline __alloc_size(1) void *kmalloc_node(size_t size, gfp_t flags, int node)
615 {
616 #ifndef CONFIG_SLOB
617 	if (__builtin_constant_p(size) &&
618 		size <= KMALLOC_MAX_CACHE_SIZE) {
619 		unsigned int i = kmalloc_index(size);
620 
621 		if (!i)
622 			return ZERO_SIZE_PTR;
623 
624 		return kmem_cache_alloc_node_trace(
625 				kmalloc_caches[kmalloc_type(flags)][i],
626 						flags, node, size);
627 	}
628 #endif
629 	return __kmalloc_node(size, flags, node);
630 }
631 
632 /**
633  * kmalloc_array - allocate memory for an array.
634  * @n: number of elements.
635  * @size: element size.
636  * @flags: the type of memory to allocate (see kmalloc).
637  */
638 static inline __alloc_size(1, 2) void *kmalloc_array(size_t n, size_t size, gfp_t flags)
639 {
640 	size_t bytes;
641 
642 	if (unlikely(check_mul_overflow(n, size, &bytes)))
643 		return NULL;
644 	if (__builtin_constant_p(n) && __builtin_constant_p(size))
645 		return kmalloc(bytes, flags);
646 	return __kmalloc(bytes, flags);
647 }
648 
649 /**
650  * krealloc_array - reallocate memory for an array.
651  * @p: pointer to the memory chunk to reallocate
652  * @new_n: new number of elements to alloc
653  * @new_size: new size of a single member of the array
654  * @flags: the type of memory to allocate (see kmalloc)
655  */
656 static inline __alloc_size(2, 3) void * __must_check krealloc_array(void *p,
657 								    size_t new_n,
658 								    size_t new_size,
659 								    gfp_t flags)
660 {
661 	size_t bytes;
662 
663 	if (unlikely(check_mul_overflow(new_n, new_size, &bytes)))
664 		return NULL;
665 
666 	return krealloc(p, bytes, flags);
667 }
668 
669 /**
670  * kcalloc - allocate memory for an array. The memory is set to zero.
671  * @n: number of elements.
672  * @size: element size.
673  * @flags: the type of memory to allocate (see kmalloc).
674  */
675 static inline __alloc_size(1, 2) void *kcalloc(size_t n, size_t size, gfp_t flags)
676 {
677 	return kmalloc_array(n, size, flags | __GFP_ZERO);
678 }
679 
680 /*
681  * kmalloc_track_caller is a special version of kmalloc that records the
682  * calling function of the routine calling it for slab leak tracking instead
683  * of just the calling function (confusing, eh?).
684  * It's useful when the call to kmalloc comes from a widely-used standard
685  * allocator where we care about the real place the memory allocation
686  * request comes from.
687  */
688 extern void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller);
689 #define kmalloc_track_caller(size, flags) \
690 	__kmalloc_track_caller(size, flags, _RET_IP_)
691 
692 static inline __alloc_size(1, 2) void *kmalloc_array_node(size_t n, size_t size, gfp_t flags,
693 							  int node)
694 {
695 	size_t bytes;
696 
697 	if (unlikely(check_mul_overflow(n, size, &bytes)))
698 		return NULL;
699 	if (__builtin_constant_p(n) && __builtin_constant_p(size))
700 		return kmalloc_node(bytes, flags, node);
701 	return __kmalloc_node(bytes, flags, node);
702 }
703 
704 static inline __alloc_size(1, 2) void *kcalloc_node(size_t n, size_t size, gfp_t flags, int node)
705 {
706 	return kmalloc_array_node(n, size, flags | __GFP_ZERO, node);
707 }
708 
709 
710 #ifdef CONFIG_NUMA
711 extern void *__kmalloc_node_track_caller(size_t size, gfp_t flags, int node,
712 					 unsigned long caller) __alloc_size(1);
713 #define kmalloc_node_track_caller(size, flags, node) \
714 	__kmalloc_node_track_caller(size, flags, node, \
715 			_RET_IP_)
716 
717 #else /* CONFIG_NUMA */
718 
719 #define kmalloc_node_track_caller(size, flags, node) \
720 	kmalloc_track_caller(size, flags)
721 
722 #endif /* CONFIG_NUMA */
723 
724 /*
725  * Shortcuts
726  */
727 static inline void *kmem_cache_zalloc(struct kmem_cache *k, gfp_t flags)
728 {
729 	return kmem_cache_alloc(k, flags | __GFP_ZERO);
730 }
731 
732 /**
733  * kzalloc - allocate memory. The memory is set to zero.
734  * @size: how many bytes of memory are required.
735  * @flags: the type of memory to allocate (see kmalloc).
736  */
737 static inline __alloc_size(1) void *kzalloc(size_t size, gfp_t flags)
738 {
739 	return kmalloc(size, flags | __GFP_ZERO);
740 }
741 
742 /**
743  * kzalloc_node - allocate zeroed memory from a particular memory node.
744  * @size: how many bytes of memory are required.
745  * @flags: the type of memory to allocate (see kmalloc).
746  * @node: memory node from which to allocate
747  */
748 static inline __alloc_size(1) void *kzalloc_node(size_t size, gfp_t flags, int node)
749 {
750 	return kmalloc_node(size, flags | __GFP_ZERO, node);
751 }
752 
753 extern void *kvmalloc_node(size_t size, gfp_t flags, int node) __alloc_size(1);
754 static inline __alloc_size(1) void *kvmalloc(size_t size, gfp_t flags)
755 {
756 	return kvmalloc_node(size, flags, NUMA_NO_NODE);
757 }
758 static inline __alloc_size(1) void *kvzalloc_node(size_t size, gfp_t flags, int node)
759 {
760 	return kvmalloc_node(size, flags | __GFP_ZERO, node);
761 }
762 static inline __alloc_size(1) void *kvzalloc(size_t size, gfp_t flags)
763 {
764 	return kvmalloc(size, flags | __GFP_ZERO);
765 }
766 
767 static inline __alloc_size(1, 2) void *kvmalloc_array(size_t n, size_t size, gfp_t flags)
768 {
769 	size_t bytes;
770 
771 	if (unlikely(check_mul_overflow(n, size, &bytes)))
772 		return NULL;
773 
774 	return kvmalloc(bytes, flags);
775 }
776 
777 static inline __alloc_size(1, 2) void *kvcalloc(size_t n, size_t size, gfp_t flags)
778 {
779 	return kvmalloc_array(n, size, flags | __GFP_ZERO);
780 }
781 
782 extern void *kvrealloc(const void *p, size_t oldsize, size_t newsize, gfp_t flags)
783 		      __alloc_size(3);
784 extern void kvfree(const void *addr);
785 extern void kvfree_sensitive(const void *addr, size_t len);
786 
787 unsigned int kmem_cache_size(struct kmem_cache *s);
788 void __init kmem_cache_init_late(void);
789 
790 #if defined(CONFIG_SMP) && defined(CONFIG_SLAB)
791 int slab_prepare_cpu(unsigned int cpu);
792 int slab_dead_cpu(unsigned int cpu);
793 #else
794 #define slab_prepare_cpu	NULL
795 #define slab_dead_cpu		NULL
796 #endif
797 
798 #endif	/* _LINUX_SLAB_H */
799