xref: /OK3568_Linux_fs/u-boot/include/memalign.h (revision 4882a59341e53eb6f0b4789bf948001014eff981)
1*4882a593Smuzhiyun /*
2*4882a593Smuzhiyun  * Copyright (c) 2015 Google, Inc
3*4882a593Smuzhiyun  *
4*4882a593Smuzhiyun  * SPDX-License-Identifier:    GPL-2.0+
5*4882a593Smuzhiyun  */
6*4882a593Smuzhiyun 
7*4882a593Smuzhiyun #ifndef __ALIGNMEM_H
8*4882a593Smuzhiyun #define __ALIGNMEM_H
9*4882a593Smuzhiyun 
10*4882a593Smuzhiyun /*
11*4882a593Smuzhiyun  * ARCH_DMA_MINALIGN is defined in asm/cache.h for each architecture.  It
12*4882a593Smuzhiyun  * is used to align DMA buffers.
13*4882a593Smuzhiyun  */
14*4882a593Smuzhiyun #ifndef __ASSEMBLY__
15*4882a593Smuzhiyun #include <asm/cache.h>
16*4882a593Smuzhiyun #include <malloc.h>
17*4882a593Smuzhiyun 
18*4882a593Smuzhiyun /*
19*4882a593Smuzhiyun  * The ALLOC_CACHE_ALIGN_BUFFER macro is used to allocate a buffer on the
20*4882a593Smuzhiyun  * stack that meets the minimum architecture alignment requirements for DMA.
21*4882a593Smuzhiyun  * Such a buffer is useful for DMA operations where flushing and invalidating
22*4882a593Smuzhiyun  * the cache before and after a read and/or write operation is required for
23*4882a593Smuzhiyun  * correct operations.
24*4882a593Smuzhiyun  *
25*4882a593Smuzhiyun  * When called the macro creates an array on the stack that is sized such
26*4882a593Smuzhiyun  * that:
27*4882a593Smuzhiyun  *
28*4882a593Smuzhiyun  * 1) The beginning of the array can be advanced enough to be aligned.
29*4882a593Smuzhiyun  *
30*4882a593Smuzhiyun  * 2) The size of the aligned portion of the array is a multiple of the minimum
31*4882a593Smuzhiyun  *    architecture alignment required for DMA.
32*4882a593Smuzhiyun  *
33*4882a593Smuzhiyun  * 3) The aligned portion contains enough space for the original number of
34*4882a593Smuzhiyun  *    elements requested.
35*4882a593Smuzhiyun  *
36*4882a593Smuzhiyun  * The macro then creates a pointer to the aligned portion of this array and
37*4882a593Smuzhiyun  * assigns to the pointer the address of the first element in the aligned
38*4882a593Smuzhiyun  * portion of the array.
39*4882a593Smuzhiyun  *
40*4882a593Smuzhiyun  * Calling the macro as:
41*4882a593Smuzhiyun  *
42*4882a593Smuzhiyun  *     ALLOC_CACHE_ALIGN_BUFFER(uint32_t, buffer, 1024);
43*4882a593Smuzhiyun  *
44*4882a593Smuzhiyun  * Will result in something similar to saying:
45*4882a593Smuzhiyun  *
46*4882a593Smuzhiyun  *     uint32_t    buffer[1024];
47*4882a593Smuzhiyun  *
48*4882a593Smuzhiyun  * The following differences exist:
49*4882a593Smuzhiyun  *
50*4882a593Smuzhiyun  * 1) The resulting buffer is guaranteed to be aligned to the value of
51*4882a593Smuzhiyun  *    ARCH_DMA_MINALIGN.
52*4882a593Smuzhiyun  *
53*4882a593Smuzhiyun  * 2) The buffer variable created by the macro is a pointer to the specified
54*4882a593Smuzhiyun  *    type, and NOT an array of the specified type.  This can be very important
55*4882a593Smuzhiyun  *    if you want the address of the buffer, which you probably do, to pass it
56*4882a593Smuzhiyun  *    to the DMA hardware.  The value of &buffer is different in the two cases.
57*4882a593Smuzhiyun  *    In the macro case it will be the address of the pointer, not the address
58*4882a593Smuzhiyun  *    of the space reserved for the buffer.  However, in the second case it
59*4882a593Smuzhiyun  *    would be the address of the buffer.  So if you are replacing hard coded
60*4882a593Smuzhiyun  *    stack buffers with this macro you need to make sure you remove the & from
61*4882a593Smuzhiyun  *    the locations where you are taking the address of the buffer.
62*4882a593Smuzhiyun  *
63*4882a593Smuzhiyun  * Note that the size parameter is the number of array elements to allocate,
64*4882a593Smuzhiyun  * not the number of bytes.
65*4882a593Smuzhiyun  *
66*4882a593Smuzhiyun  * This macro can not be used outside of function scope, or for the creation
67*4882a593Smuzhiyun  * of a function scoped static buffer.  It can not be used to create a cache
68*4882a593Smuzhiyun  * line aligned global buffer.
69*4882a593Smuzhiyun  */
70*4882a593Smuzhiyun #define PAD_COUNT(s, pad) (((s) - 1) / (pad) + 1)
71*4882a593Smuzhiyun #define PAD_SIZE(s, pad) (PAD_COUNT(s, pad) * pad)
72*4882a593Smuzhiyun #define ALLOC_ALIGN_BUFFER_PAD(type, name, size, align, pad)		\
73*4882a593Smuzhiyun 	char __##name[ROUND(PAD_SIZE((size) * sizeof(type), pad), align)  \
74*4882a593Smuzhiyun 		      + (align - 1)];					\
75*4882a593Smuzhiyun 									\
76*4882a593Smuzhiyun 	type *name = (type *)ALIGN((uintptr_t)__##name, align)
77*4882a593Smuzhiyun #define ALLOC_ALIGN_BUFFER(type, name, size, align)		\
78*4882a593Smuzhiyun 	ALLOC_ALIGN_BUFFER_PAD(type, name, size, align, 1)
79*4882a593Smuzhiyun #define ALLOC_CACHE_ALIGN_BUFFER_PAD(type, name, size, pad)		\
80*4882a593Smuzhiyun 	ALLOC_ALIGN_BUFFER_PAD(type, name, size, ARCH_DMA_MINALIGN, pad)
81*4882a593Smuzhiyun #define ALLOC_CACHE_ALIGN_BUFFER(type, name, size)			\
82*4882a593Smuzhiyun 	ALLOC_ALIGN_BUFFER(type, name, size, ARCH_DMA_MINALIGN)
83*4882a593Smuzhiyun 
84*4882a593Smuzhiyun /*
85*4882a593Smuzhiyun  * DEFINE_CACHE_ALIGN_BUFFER() is similar to ALLOC_CACHE_ALIGN_BUFFER, but it's
86*4882a593Smuzhiyun  * purpose is to allow allocating aligned buffers outside of function scope.
87*4882a593Smuzhiyun  * Usage of this macro shall be avoided or used with extreme care!
88*4882a593Smuzhiyun  */
89*4882a593Smuzhiyun #define DEFINE_ALIGN_BUFFER(type, name, size, align)			\
90*4882a593Smuzhiyun 	static char __##name[ALIGN(size * sizeof(type), align)]	\
91*4882a593Smuzhiyun 			__aligned(align);				\
92*4882a593Smuzhiyun 									\
93*4882a593Smuzhiyun 	static type *name = (type *)__##name
94*4882a593Smuzhiyun #define DEFINE_CACHE_ALIGN_BUFFER(type, name, size)			\
95*4882a593Smuzhiyun 	DEFINE_ALIGN_BUFFER(type, name, size, ARCH_DMA_MINALIGN)
96*4882a593Smuzhiyun 
97*4882a593Smuzhiyun /**
98*4882a593Smuzhiyun  * malloc_cache_aligned() - allocate a memory region aligned to cache line size
99*4882a593Smuzhiyun  *
100*4882a593Smuzhiyun  * This allocates memory at a cache-line boundary. The amount allocated may
101*4882a593Smuzhiyun  * be larger than requested as it is rounded up to the nearest multiple of the
102*4882a593Smuzhiyun  * cache-line size. This ensured that subsequent cache operations on this
103*4882a593Smuzhiyun  * memory (flush, invalidate) will not affect subsequently allocated regions.
104*4882a593Smuzhiyun  *
105*4882a593Smuzhiyun  * @size:	Minimum number of bytes to allocate
106*4882a593Smuzhiyun  *
107*4882a593Smuzhiyun  * @return pointer to new memory region, or NULL if there is no more memory
108*4882a593Smuzhiyun  * available.
109*4882a593Smuzhiyun  */
malloc_cache_aligned(size_t size)110*4882a593Smuzhiyun static inline void *malloc_cache_aligned(size_t size)
111*4882a593Smuzhiyun {
112*4882a593Smuzhiyun 	return memalign(ARCH_DMA_MINALIGN, ALIGN(size, ARCH_DMA_MINALIGN));
113*4882a593Smuzhiyun }
114*4882a593Smuzhiyun #endif
115*4882a593Smuzhiyun 
116*4882a593Smuzhiyun #endif /* __ALIGNMEM_H */
117