Gitiles
Code Review Sign In
gerrit-public.fairphone.software / kernel / msm-4.19 / a81d72d2002d6a932bd83022cbf8c442b1b97512 / . / arch / x86 / lib / memset_64.S
blob: 55b95db30a61c08df145e2c321fb0362696a623f [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -0700[diff] [blame]1/* Copyright 2002 Andi Kleen, SuSE Labs */
Jan Beulich8d379da2006-09-26 10:52:32 +0200[diff] [blame]2
Jan Beulich8d379da2006-09-26 10:52:32 +0200[diff] [blame]3#include <linux/linkage.h>
Borislav Petkovcd4d09e2016-01-26 22:12:04 +0100[diff] [blame]4#include <asm/cpufeatures.h>
Fenghua Yu2f19e062011-05-17 15:29:18 -0700[diff] [blame]5#include <asm/alternative-asm.h>
Al Viro784d5692016-01-11 11:04:34 -0500[diff] [blame]6#include <asm/export.h>
Jan Beulich8d379da2006-09-26 10:52:32 +0200[diff] [blame]7
Borislav Petkov84d95ad2015-02-04 08:57:00 +0100[diff] [blame]8.weak memset
9
Linus Torvalds1da177e2005-04-16 15:20:36 -0700[diff] [blame]10/*
Fenghua Yu2f19e062011-05-17 15:29:18 -0700[diff] [blame]11 * 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
Adam Buchbinder6a6256f2016-02-23 15:34:30 -0800[diff] [blame]13 * simpler and shorter than the original function as well.
Borislav Petkov84d95ad2015-02-04 08:57:00 +0100[diff] [blame]14 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700[diff] [blame]15 * rdi destination
Borislav Petkov84d95ad2015-02-04 08:57:00 +0100[diff] [blame]16 * rsi value (char)
17 * rdx count (bytes)
18 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700[diff] [blame]19 * rax original destination
Borislav Petkov84d95ad2015-02-04 08:57:00 +0100[diff] [blame]20 */
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 +0200[diff] [blame]32 movq %rdi,%r9
Jan Beulich5d7244e2012-01-05 16:10:42 +0000[diff] [blame]33 movq %rdx,%rcx
34 andl $7,%edx
35 shrq $3,%rcx
Jan Beulich8d379da2006-09-26 10:52:32 +0200[diff] [blame]36 /* expand byte value */
37 movzbl %sil,%esi
38 movabs $0x0101010101010101,%rax
Jan Beulich5d7244e2012-01-05 16:10:42 +0000[diff] [blame]39 imulq %rsi,%rax
Jan Beulich8d379da2006-09-26 10:52:32 +0200[diff] [blame]40 rep stosq
Jan Beulich5d7244e2012-01-05 16:10:42 +0000[diff] [blame]41 movl %edx,%ecx
Jan Beulich8d379da2006-09-26 10:52:32 +0200[diff] [blame]42 rep stosb
43 movq %r9,%rax
44 ret
Borislav Petkov84d95ad2015-02-04 08:57:00 +0100[diff] [blame]45ENDPROC(memset)
46ENDPROC(__memset)
Al Viro784d5692016-01-11 11:04:34 -0500[diff] [blame]47EXPORT_SYMBOL(memset)
48EXPORT_SYMBOL(__memset)
Jan Beulich8d379da2006-09-26 10:52:32 +0200[diff] [blame]49
Fenghua Yu2f19e062011-05-17 15:29:18 -0700[diff] [blame]50/*
51 * ISO C memset - set a memory block to a byte value. This function uses
52 * enhanced rep stosb to override the fast string function.
53 * The code is simpler and shorter than the fast string function as well.
54 *
55 * rdi destination
56 * rsi value (char)
57 * rdx count (bytes)
58 *
59 * rax original destination
60 */
Borislav Petkov84d95ad2015-02-04 08:57:00 +0100[diff] [blame]61ENTRY(memset_erms)
Fenghua Yu2f19e062011-05-17 15:29:18 -0700[diff] [blame]62 movq %rdi,%r9
63 movb %sil,%al
Jan Beulich5d7244e2012-01-05 16:10:42 +0000[diff] [blame]64 movq %rdx,%rcx
Fenghua Yu2f19e062011-05-17 15:29:18 -0700[diff] [blame]65 rep stosb
66 movq %r9,%rax
67 ret
Borislav Petkov84d95ad2015-02-04 08:57:00 +0100[diff] [blame]68ENDPROC(memset_erms)
Fenghua Yu2f19e062011-05-17 15:29:18 -0700[diff] [blame]69
Borislav Petkov84d95ad2015-02-04 08:57:00 +0100[diff] [blame]70ENTRY(memset_orig)
Andi Kleen7bcd3f32006-02-03 21:51:02 +0100[diff] [blame]71 movq %rdi,%r10
Andi Kleen7bcd3f32006-02-03 21:51:02 +0100[diff] [blame]72
73 /* expand byte value */
74 movzbl %sil,%ecx
75 movabs $0x0101010101010101,%rax
Jan Beulich5d7244e2012-01-05 16:10:42 +0000[diff] [blame]76 imulq %rcx,%rax
Andi Kleen7bcd3f32006-02-03 21:51:02 +0100[diff] [blame]77
78 /* align dst */
79 movl %edi,%r9d
80 andl $7,%r9d
81 jnz .Lbad_alignment
82.Lafter_bad_alignment:
83
Jan Beulich5d7244e2012-01-05 16:10:42 +0000[diff] [blame]84 movq %rdx,%rcx
85 shrq $6,%rcx
Andi Kleen7bcd3f32006-02-03 21:51:02 +0100[diff] [blame]86 jz .Lhandle_tail
87
88 .p2align 4
89.Lloop_64:
Jan Beulich5d7244e2012-01-05 16:10:42 +0000[diff] [blame]90 decq %rcx
Andi Kleen7bcd3f32006-02-03 21:51:02 +0100[diff] [blame]91 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 +0000[diff] [blame]106 movl %edx,%ecx
Andi Kleen7bcd3f32006-02-03 21:51:02 +0100[diff] [blame]107 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 +0000[diff] [blame]118 andl $7,%edx
Andi Kleen7bcd3f32006-02-03 21:51:02 +0100[diff] [blame]119 jz .Lende
120 .p2align 4
121.Lloop_1:
Jan Beulich5d7244e2012-01-05 16:10:42 +0000[diff] [blame]122 decl %edx
Andi Kleen7bcd3f32006-02-03 21:51:02 +0100[diff] [blame]123 movb %al,(%rdi)
124 leaq 1(%rdi),%rdi
125 jnz .Lloop_1
126
127.Lende:
128 movq %r10,%rax
129 ret
130
131.Lbad_alignment:
Jan Beulich5d7244e2012-01-05 16:10:42 +0000[diff] [blame]132 cmpq $7,%rdx
Andi Kleen7bcd3f32006-02-03 21:51:02 +0100[diff] [blame]133 jbe .Lhandle_7
134 movq %rax,(%rdi) /* unaligned store */
135 movq $8,%r8
136 subq %r9,%r8
137 addq %r8,%rdi
Jan Beulich5d7244e2012-01-05 16:10:42 +0000[diff] [blame]138 subq %r8,%rdx
Andi Kleen7bcd3f32006-02-03 21:51:02 +0100[diff] [blame]139 jmp .Lafter_bad_alignment
Jan Beulich8d379da2006-09-26 10:52:32 +0200[diff] [blame]140.Lfinal:
Borislav Petkov84d95ad2015-02-04 08:57:00 +0100[diff] [blame]141ENDPROC(memset_orig)