blob: 93118fb239762ba78efd754d20a756aed0221411 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/* Copyright 2002 Andi Kleen, SuSE Labs */
Jan Beulich8d379da2006-09-26 10:52:32 +02002
Jan Beulich8d379da2006-09-26 10:52:32 +02003#include <linux/linkage.h>
4#include <asm/dwarf2.h>
Fenghua Yu2f19e062011-05-17 15:29:18 -07005#include <asm/cpufeature.h>
6#include <asm/alternative-asm.h>
Jan Beulich8d379da2006-09-26 10:52:32 +02007
Borislav Petkov84d95ad2015-02-04 08:57:00 +01008.weak memset
9
Linus Torvalds1da177e2005-04-16 15:20:36 -070010/*
Fenghua Yu2f19e062011-05-17 15:29:18 -070011 * ISO C memset - set a memory block to a byte value. This function uses fast
12 * string to get better performance than the original function. The code is
13 * simpler and shorter than the orignal function as well.
Borislav Petkov84d95ad2015-02-04 08:57:00 +010014 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070015 * rdi destination
Borislav Petkov84d95ad2015-02-04 08:57:00 +010016 * rsi value (char)
17 * rdx count (bytes)
18 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070019 * rax original destination
Borislav Petkov84d95ad2015-02-04 08:57:00 +010020 */
21ENTRY(memset)
22ENTRY(__memset)
23 /*
24 * Some CPUs support enhanced REP MOVSB/STOSB feature. It is recommended
25 * to use it when possible. If not available, use fast string instructions.
26 *
27 * Otherwise, use original memset function.
28 */
29 ALTERNATIVE_2 "jmp memset_orig", "", X86_FEATURE_REP_GOOD, \
30 "jmp memset_erms", X86_FEATURE_ERMS
31
Jan Beulich8d379da2006-09-26 10:52:32 +020032 movq %rdi,%r9
Jan Beulich5d7244e2012-01-05 16:10:42 +000033 movq %rdx,%rcx
34 andl $7,%edx
35 shrq $3,%rcx
Jan Beulich8d379da2006-09-26 10:52:32 +020036 /* expand byte value */
37 movzbl %sil,%esi
38 movabs $0x0101010101010101,%rax
Jan Beulich5d7244e2012-01-05 16:10:42 +000039 imulq %rsi,%rax
Jan Beulich8d379da2006-09-26 10:52:32 +020040 rep stosq
Jan Beulich5d7244e2012-01-05 16:10:42 +000041 movl %edx,%ecx
Jan Beulich8d379da2006-09-26 10:52:32 +020042 rep stosb
43 movq %r9,%rax
44 ret
Borislav Petkov84d95ad2015-02-04 08:57:00 +010045ENDPROC(memset)
46ENDPROC(__memset)
Jan Beulich8d379da2006-09-26 10:52:32 +020047
Fenghua Yu2f19e062011-05-17 15:29:18 -070048/*
49 * ISO C memset - set a memory block to a byte value. This function uses
50 * enhanced rep stosb to override the fast string function.
51 * The code is simpler and shorter than the fast string function as well.
52 *
53 * rdi destination
54 * rsi value (char)
55 * rdx count (bytes)
56 *
57 * rax original destination
58 */
Borislav Petkov84d95ad2015-02-04 08:57:00 +010059ENTRY(memset_erms)
Fenghua Yu2f19e062011-05-17 15:29:18 -070060 movq %rdi,%r9
61 movb %sil,%al
Jan Beulich5d7244e2012-01-05 16:10:42 +000062 movq %rdx,%rcx
Fenghua Yu2f19e062011-05-17 15:29:18 -070063 rep stosb
64 movq %r9,%rax
65 ret
Borislav Petkov84d95ad2015-02-04 08:57:00 +010066ENDPROC(memset_erms)
Fenghua Yu2f19e062011-05-17 15:29:18 -070067
Borislav Petkov84d95ad2015-02-04 08:57:00 +010068ENTRY(memset_orig)
Jan Beulich8d379da2006-09-26 10:52:32 +020069 CFI_STARTPROC
Andi Kleen7bcd3f32006-02-03 21:51:02 +010070 movq %rdi,%r10
Andi Kleen7bcd3f32006-02-03 21:51:02 +010071
72 /* expand byte value */
73 movzbl %sil,%ecx
74 movabs $0x0101010101010101,%rax
Jan Beulich5d7244e2012-01-05 16:10:42 +000075 imulq %rcx,%rax
Andi Kleen7bcd3f32006-02-03 21:51:02 +010076
77 /* align dst */
78 movl %edi,%r9d
79 andl $7,%r9d
80 jnz .Lbad_alignment
Jan Beulich8d379da2006-09-26 10:52:32 +020081 CFI_REMEMBER_STATE
Andi Kleen7bcd3f32006-02-03 21:51:02 +010082.Lafter_bad_alignment:
83
Jan Beulich5d7244e2012-01-05 16:10:42 +000084 movq %rdx,%rcx
85 shrq $6,%rcx
Andi Kleen7bcd3f32006-02-03 21:51:02 +010086 jz .Lhandle_tail
87
88 .p2align 4
89.Lloop_64:
Jan Beulich5d7244e2012-01-05 16:10:42 +000090 decq %rcx
Andi Kleen7bcd3f32006-02-03 21:51:02 +010091 movq %rax,(%rdi)
92 movq %rax,8(%rdi)
93 movq %rax,16(%rdi)
94 movq %rax,24(%rdi)
95 movq %rax,32(%rdi)
96 movq %rax,40(%rdi)
97 movq %rax,48(%rdi)
98 movq %rax,56(%rdi)
99 leaq 64(%rdi),%rdi
100 jnz .Lloop_64
101
102 /* Handle tail in loops. The loops should be faster than hard
103 to predict jump tables. */
104 .p2align 4
105.Lhandle_tail:
Jan Beulich5d7244e2012-01-05 16:10:42 +0000106 movl %edx,%ecx
Andi Kleen7bcd3f32006-02-03 21:51:02 +0100107 andl $63&(~7),%ecx
108 jz .Lhandle_7
109 shrl $3,%ecx
110 .p2align 4
111.Lloop_8:
112 decl %ecx
113 movq %rax,(%rdi)
114 leaq 8(%rdi),%rdi
115 jnz .Lloop_8
116
117.Lhandle_7:
Jan Beulich5d7244e2012-01-05 16:10:42 +0000118 andl $7,%edx
Andi Kleen7bcd3f32006-02-03 21:51:02 +0100119 jz .Lende
120 .p2align 4
121.Lloop_1:
Jan Beulich5d7244e2012-01-05 16:10:42 +0000122 decl %edx
Andi Kleen7bcd3f32006-02-03 21:51:02 +0100123 movb %al,(%rdi)
124 leaq 1(%rdi),%rdi
125 jnz .Lloop_1
126
127.Lende:
128 movq %r10,%rax
129 ret
130
Jan Beulich8d379da2006-09-26 10:52:32 +0200131 CFI_RESTORE_STATE
Andi Kleen7bcd3f32006-02-03 21:51:02 +0100132.Lbad_alignment:
Jan Beulich5d7244e2012-01-05 16:10:42 +0000133 cmpq $7,%rdx
Andi Kleen7bcd3f32006-02-03 21:51:02 +0100134 jbe .Lhandle_7
135 movq %rax,(%rdi) /* unaligned store */
136 movq $8,%r8
137 subq %r9,%r8
138 addq %r8,%rdi
Jan Beulich5d7244e2012-01-05 16:10:42 +0000139 subq %r8,%rdx
Andi Kleen7bcd3f32006-02-03 21:51:02 +0100140 jmp .Lafter_bad_alignment
Jan Beulich8d379da2006-09-26 10:52:32 +0200141.Lfinal:
142 CFI_ENDPROC
Borislav Petkov84d95ad2015-02-04 08:57:00 +0100143ENDPROC(memset_orig)