xref: /openbmc/linux/arch/x86/lib/memset_64.S (revision e2c75e76)
1/* SPDX-License-Identifier: GPL-2.0 */
2/* Copyright 2002 Andi Kleen, SuSE Labs */
3
4#include <linux/linkage.h>
5#include <asm/cpufeatures.h>
6#include <asm/alternative-asm.h>
7#include <asm/export.h>
8
9.weak memset
10
11/*
12 * ISO C memset - set a memory block to a byte value. This function uses fast
13 * string to get better performance than the original function. The code is
14 * simpler and shorter than the original function as well.
15 *
16 * rdi   destination
17 * rsi   value (char)
18 * rdx   count (bytes)
19 *
20 * rax   original destination
21 */
22ENTRY(memset)
23ENTRY(__memset)
24	/*
25	 * Some CPUs support enhanced REP MOVSB/STOSB feature. It is recommended
26	 * to use it when possible. If not available, use fast string instructions.
27	 *
28	 * Otherwise, use original memset function.
29	 */
30	ALTERNATIVE_2 "jmp memset_orig", "", X86_FEATURE_REP_GOOD, \
31		      "jmp memset_erms", X86_FEATURE_ERMS
32
33	movq %rdi,%r9
34	movq %rdx,%rcx
35	andl $7,%edx
36	shrq $3,%rcx
37	/* expand byte value  */
38	movzbl %sil,%esi
39	movabs $0x0101010101010101,%rax
40	imulq %rsi,%rax
41	rep stosq
42	movl %edx,%ecx
43	rep stosb
44	movq %r9,%rax
45	ret
46ENDPROC(memset)
47ENDPROC(__memset)
48EXPORT_SYMBOL(memset)
49EXPORT_SYMBOL(__memset)
50
51/*
52 * ISO C memset - set a memory block to a byte value. This function uses
53 * enhanced rep stosb to override the fast string function.
54 * The code is simpler and shorter than the fast string function as well.
55 *
56 * rdi   destination
57 * rsi   value (char)
58 * rdx   count (bytes)
59 *
60 * rax   original destination
61 */
62ENTRY(memset_erms)
63	movq %rdi,%r9
64	movb %sil,%al
65	movq %rdx,%rcx
66	rep stosb
67	movq %r9,%rax
68	ret
69ENDPROC(memset_erms)
70
71ENTRY(memset_orig)
72	movq %rdi,%r10
73
74	/* expand byte value  */
75	movzbl %sil,%ecx
76	movabs $0x0101010101010101,%rax
77	imulq  %rcx,%rax
78
79	/* align dst */
80	movl  %edi,%r9d
81	andl  $7,%r9d
82	jnz  .Lbad_alignment
83.Lafter_bad_alignment:
84
85	movq  %rdx,%rcx
86	shrq  $6,%rcx
87	jz	 .Lhandle_tail
88
89	.p2align 4
90.Lloop_64:
91	decq  %rcx
92	movq  %rax,(%rdi)
93	movq  %rax,8(%rdi)
94	movq  %rax,16(%rdi)
95	movq  %rax,24(%rdi)
96	movq  %rax,32(%rdi)
97	movq  %rax,40(%rdi)
98	movq  %rax,48(%rdi)
99	movq  %rax,56(%rdi)
100	leaq  64(%rdi),%rdi
101	jnz    .Lloop_64
102
103	/* Handle tail in loops. The loops should be faster than hard
104	   to predict jump tables. */
105	.p2align 4
106.Lhandle_tail:
107	movl	%edx,%ecx
108	andl    $63&(~7),%ecx
109	jz 		.Lhandle_7
110	shrl	$3,%ecx
111	.p2align 4
112.Lloop_8:
113	decl   %ecx
114	movq  %rax,(%rdi)
115	leaq  8(%rdi),%rdi
116	jnz    .Lloop_8
117
118.Lhandle_7:
119	andl	$7,%edx
120	jz      .Lende
121	.p2align 4
122.Lloop_1:
123	decl    %edx
124	movb 	%al,(%rdi)
125	leaq	1(%rdi),%rdi
126	jnz     .Lloop_1
127
128.Lende:
129	movq	%r10,%rax
130	ret
131
132.Lbad_alignment:
133	cmpq $7,%rdx
134	jbe	.Lhandle_7
135	movq %rax,(%rdi)	/* unaligned store */
136	movq $8,%r8
137	subq %r9,%r8
138	addq %r8,%rdi
139	subq %r8,%rdx
140	jmp .Lafter_bad_alignment
141.Lfinal:
142ENDPROC(memset_orig)
143