1*4882a593Smuzhiyun // SPDX-License-Identifier: GPL-2.0-only
2*4882a593Smuzhiyun /*
3*4882a593Smuzhiyun * linux/lib/raid6/neon.c - RAID6 syndrome calculation using ARM NEON intrinsics
4*4882a593Smuzhiyun *
5*4882a593Smuzhiyun * Copyright (C) 2013 Linaro Ltd <ard.biesheuvel@linaro.org>
6*4882a593Smuzhiyun */
7*4882a593Smuzhiyun
8*4882a593Smuzhiyun #include <linux/raid/pq.h>
9*4882a593Smuzhiyun
10*4882a593Smuzhiyun #ifdef __KERNEL__
11*4882a593Smuzhiyun #include <asm/neon.h>
12*4882a593Smuzhiyun #else
13*4882a593Smuzhiyun #define kernel_neon_begin()
14*4882a593Smuzhiyun #define kernel_neon_end()
15*4882a593Smuzhiyun #define cpu_has_neon() (1)
16*4882a593Smuzhiyun #endif
17*4882a593Smuzhiyun
18*4882a593Smuzhiyun /*
19*4882a593Smuzhiyun * There are 2 reasons these wrappers are kept in a separate compilation unit
20*4882a593Smuzhiyun * from the actual implementations in neonN.c (generated from neon.uc by
21*4882a593Smuzhiyun * unroll.awk):
22*4882a593Smuzhiyun * - the actual implementations use NEON intrinsics, and the GCC support header
23*4882a593Smuzhiyun * (arm_neon.h) is not fully compatible (type wise) with the kernel;
24*4882a593Smuzhiyun * - the neonN.c files are compiled with -mfpu=neon and optimization enabled,
25*4882a593Smuzhiyun * and we have to make sure that we never use *any* NEON/VFP instructions
26*4882a593Smuzhiyun * outside a kernel_neon_begin()/kernel_neon_end() pair.
27*4882a593Smuzhiyun */
28*4882a593Smuzhiyun
29*4882a593Smuzhiyun #define RAID6_NEON_WRAPPER(_n) \
30*4882a593Smuzhiyun static void raid6_neon ## _n ## _gen_syndrome(int disks, \
31*4882a593Smuzhiyun size_t bytes, void **ptrs) \
32*4882a593Smuzhiyun { \
33*4882a593Smuzhiyun void raid6_neon ## _n ## _gen_syndrome_real(int, \
34*4882a593Smuzhiyun unsigned long, void**); \
35*4882a593Smuzhiyun kernel_neon_begin(); \
36*4882a593Smuzhiyun raid6_neon ## _n ## _gen_syndrome_real(disks, \
37*4882a593Smuzhiyun (unsigned long)bytes, ptrs); \
38*4882a593Smuzhiyun kernel_neon_end(); \
39*4882a593Smuzhiyun } \
40*4882a593Smuzhiyun static void raid6_neon ## _n ## _xor_syndrome(int disks, \
41*4882a593Smuzhiyun int start, int stop, \
42*4882a593Smuzhiyun size_t bytes, void **ptrs) \
43*4882a593Smuzhiyun { \
44*4882a593Smuzhiyun void raid6_neon ## _n ## _xor_syndrome_real(int, \
45*4882a593Smuzhiyun int, int, unsigned long, void**); \
46*4882a593Smuzhiyun kernel_neon_begin(); \
47*4882a593Smuzhiyun raid6_neon ## _n ## _xor_syndrome_real(disks, \
48*4882a593Smuzhiyun start, stop, (unsigned long)bytes, ptrs); \
49*4882a593Smuzhiyun kernel_neon_end(); \
50*4882a593Smuzhiyun } \
51*4882a593Smuzhiyun struct raid6_calls const raid6_neonx ## _n = { \
52*4882a593Smuzhiyun raid6_neon ## _n ## _gen_syndrome, \
53*4882a593Smuzhiyun raid6_neon ## _n ## _xor_syndrome, \
54*4882a593Smuzhiyun raid6_have_neon, \
55*4882a593Smuzhiyun "neonx" #_n, \
56*4882a593Smuzhiyun 0 \
57*4882a593Smuzhiyun }
58*4882a593Smuzhiyun
raid6_have_neon(void)59*4882a593Smuzhiyun static int raid6_have_neon(void)
60*4882a593Smuzhiyun {
61*4882a593Smuzhiyun return cpu_has_neon();
62*4882a593Smuzhiyun }
63*4882a593Smuzhiyun
64*4882a593Smuzhiyun RAID6_NEON_WRAPPER(1);
65*4882a593Smuzhiyun RAID6_NEON_WRAPPER(2);
66*4882a593Smuzhiyun RAID6_NEON_WRAPPER(4);
67*4882a593Smuzhiyun RAID6_NEON_WRAPPER(8);
68