1 /* SPDX-License-Identifier: GPL-2.0-only */
3 * Copyright (C) 2013 ARM Ltd.
4 * Copyright (C) 2013 Linaro.
6 * This code is based on glibc cortex strings work originally authored by Linaro
9 * http://bazaar.launchpad.net/~linaro-toolchain-dev/cortex-strings/trunk/
10 * files/head:/src/aarch64/
13 #include <linux/linkage.h>
14 #include <asm/assembler.h>
15 #include <asm/cache.h>
18 * Fill in the buffer with character c (alignment handled by the hardware)
45 SYM_FUNC_START_ALIAS(__memset)
46 SYM_FUNC_START_WEAK_PI(memset)
47 mov dst, dstin /* Preserve return value. */
49 orr A_lw, A_lw, A_lw, lsl #8
50 orr A_lw, A_lw, A_lw, lsl #16
51 orr A_l, A_l, A_l, lsl #32
55 /*All store maybe are non-aligned..*/
71 /*Whether the start address is aligned with 16.*/
76 * The count is not less than 16, we can use stp to store the start 16 bytes,
77 * then adjust the dst aligned with 16.This process will make the current
78 * memory address at alignment boundary.
80 stp A_l, A_l, [dst] /*non-aligned store..*/
81 /*make the dst aligned..*/
82 sub count, count, tmp2
92 ands tmp1, count, #0x30
97 stp A_l, A_l, [dst], #16
99 stp A_l, A_l, [dst], #16
101 stp A_l, A_l, [dst], #16
103 * The last store length is less than 16,use stp to write last 16 bytes.
104 * It will lead some bytes written twice and the access is non-aligned.
107 ands count, count, #15
110 stp A_l, A_l, [dst, #-16] /* Repeat some/all of last store. */
115 * Critical loop. Start at a new cache line boundary. Assuming
116 * 64 bytes per line, this ensures the entire loop is in one line.
118 .p2align L1_CACHE_SHIFT
120 sub dst, dst, #16/* Pre-bias. */
121 sub count, count, #64
123 stp A_l, A_l, [dst, #16]
124 stp A_l, A_l, [dst, #32]
125 stp A_l, A_l, [dst, #48]
126 stp A_l, A_l, [dst, #64]!
127 subs count, count, #64
136 * For zeroing memory, check to see if we can use the ZVA feature to
137 * zero entire 'cache' lines.
143 * For zeroing small amounts of memory, it's not worth setting up
144 * the line-clear code.
147 b.lt .Lnot_short /*count is at least 128 bytes*/
150 tbnz tmp1, #4, .Lnot_short
152 and zva_len, tmp1w, #15 /* Safety: other bits reserved. */
153 lsl zva_len, tmp3w, zva_len
155 ands tmp3w, zva_len, #63
157 * ensure the zva_len is not less than 64.
158 * It is not meaningful to use ZVA if the block size is less than 64.
163 * Compute how far we need to go to become suitably aligned. We're
164 * already at quad-word alignment.
167 b.lt .Lnot_short /* Not enough to reach alignment. */
168 sub zva_bits_x, zva_len_x, #1
170 ands tmp2, tmp2, zva_bits_x
171 b.eq 2f /* Already aligned. */
172 /* Not aligned, check that there's enough to copy after alignment.*/
173 sub tmp1, count, tmp2
175 * grantee the remain length to be ZVA is bigger than 64,
176 * avoid to make the 2f's process over mem range.*/
178 ccmp tmp1, zva_len_x, #8, ge /* NZCV=0b1000 */
181 * We know that there's at least 64 bytes to zero and that it's safe
182 * to overrun by 64 bytes.
187 stp A_l, A_l, [dst, #16]
188 stp A_l, A_l, [dst, #32]
190 stp A_l, A_l, [dst, #48]
193 /* We've overrun a bit, so adjust dst downwards.*/
196 sub count, count, zva_len_x
199 add dst, dst, zva_len_x
200 subs count, count, zva_len_x
202 ands count, count, zva_bits_x
203 b.ne .Ltail_maybe_long
205 SYM_FUNC_END_PI(memset)
206 EXPORT_SYMBOL(memset)
207 SYM_FUNC_END_ALIAS(__memset)
208 EXPORT_SYMBOL(__memset)