1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Normally compiler builtins are used, but sometimes the compiler calls out
4 * of line code. Based on asm-i386/string.h.
6 * This assembly file is re-written from memmove_64.c file.
7 * - Copyright 2011 Fenghua Yu <fenghua.yu@intel.com>
9 #include <linux/linkage.h>
10 #include <asm/cpufeatures.h>
11 #include <asm/alternative-asm.h>
12 #include <asm/export.h>
17 * Implement memmove(). This can handle overlap between src and dst.
32 /* Handle more 32 bytes in loop */
37 /* Decide forward/backward copy mode */
39 jge .Lmemmove_begin_forward
45 .Lmemmove_begin_forward:
46 ALTERNATIVE "", "movq %rdx, %rcx; rep movsb; retq", X86_FEATURE_ERMS
49 * movsq instruction have many startup latency
50 * so we handle small size by general register.
55 * movsq instruction is only good for aligned case.
63 * We gobble 32 bytes forward in each loop.
82 * Handle data forward by movsq.
87 movq -8(%rsi, %rdx), %r11
88 lea -8(%rdi, %rdx), %r10
93 .Lmemmove_end_forward:
96 * Handle data backward by movsq.
103 leaq -8(%rsi, %rdx), %rsi
104 leaq -8(%rdi, %rdx), %rdi
113 * Start to prepare for backward copy.
123 * Calculate copy position to tail.
129 * We gobble 32 bytes backward in each loop.
133 movq -1*8(%rsi), %r11
134 movq -2*8(%rsi), %r10
137 leaq -4*8(%rsi), %rsi
139 movq %r11, -1*8(%rdi)
140 movq %r10, -2*8(%rdi)
143 leaq -4*8(%rdi), %rdi
146 * Calculate copy position to head.
155 * Move data from 16 bytes to 31 bytes.
159 movq -2*8(%rsi, %rdx), %r9
160 movq -1*8(%rsi, %rdx), %r8
163 movq %r9, -2*8(%rdi, %rdx)
164 movq %r8, -1*8(%rdi, %rdx)
171 * Move data from 8 bytes to 15 bytes.
174 movq -1*8(%rsi, %rdx), %r10
176 movq %r10, -1*8(%rdi, %rdx)
182 * Move data from 4 bytes to 7 bytes.
185 movl -4(%rsi, %rdx), %r10d
187 movl %r10d, -4(%rdi, %rdx)
193 * Move data from 2 bytes to 3 bytes.
196 movw -2(%rsi, %rdx), %r10w
198 movw %r10w, -2(%rdi, %rdx)
204 * Move data for 1 byte.
212 EXPORT_SYMBOL(__memmove)
213 EXPORT_SYMBOL(memmove)