aboutsummaryrefslogtreecommitdiff
path: root/arch/x86/lib/memset_64.S
diff options
context:
space:
mode:
Diffstat (limited to 'arch/x86/lib/memset_64.S')
-rw-r--r--arch/x86/lib/memset_64.S85
1 files changed, 56 insertions, 29 deletions
diff --git a/arch/x86/lib/memset_64.S b/arch/x86/lib/memset_64.S
index 09d34426965..2dcb3808cbd 100644
--- a/arch/x86/lib/memset_64.S
+++ b/arch/x86/lib/memset_64.S
@@ -2,9 +2,13 @@
#include <linux/linkage.h>
#include <asm/dwarf2.h>
+#include <asm/cpufeature.h>
+#include <asm/alternative-asm.h>
/*
- * ISO C memset - set a memory block to a byte value.
+ * ISO C memset - set a memory block to a byte value. This function uses fast
+ * string to get better performance than the original function. The code is
+ * simpler and shorter than the orignal function as well.
*
* rdi destination
* rsi value (char)
@@ -15,32 +19,52 @@
.section .altinstr_replacement, "ax", @progbits
.Lmemset_c:
movq %rdi,%r9
- movl %edx,%r8d
- andl $7,%r8d
- movl %edx,%ecx
- shrl $3,%ecx
+ movq %rdx,%rcx
+ andl $7,%edx
+ shrq $3,%rcx
/* expand byte value */
movzbl %sil,%esi
movabs $0x0101010101010101,%rax
- mulq %rsi /* with rax, clobbers rdx */
+ imulq %rsi,%rax
rep stosq
- movl %r8d,%ecx
+ movl %edx,%ecx
rep stosb
movq %r9,%rax
ret
.Lmemset_e:
.previous
+/*
+ * ISO C memset - set a memory block to a byte value. This function uses
+ * enhanced rep stosb to override the fast string function.
+ * The code is simpler and shorter than the fast string function as well.
+ *
+ * rdi destination
+ * rsi value (char)
+ * rdx count (bytes)
+ *
+ * rax original destination
+ */
+ .section .altinstr_replacement, "ax", @progbits
+.Lmemset_c_e:
+ movq %rdi,%r9
+ movb %sil,%al
+ movq %rdx,%rcx
+ rep stosb
+ movq %r9,%rax
+ ret
+.Lmemset_e_e:
+ .previous
+
ENTRY(memset)
ENTRY(__memset)
CFI_STARTPROC
movq %rdi,%r10
- movq %rdx,%r11
/* expand byte value */
movzbl %sil,%ecx
movabs $0x0101010101010101,%rax
- mul %rcx /* with rax, clobbers rdx */
+ imulq %rcx,%rax
/* align dst */
movl %edi,%r9d
@@ -49,13 +73,13 @@ ENTRY(__memset)
CFI_REMEMBER_STATE
.Lafter_bad_alignment:
- movl %r11d,%ecx
- shrl $6,%ecx
+ movq %rdx,%rcx
+ shrq $6,%rcx
jz .Lhandle_tail
.p2align 4
.Lloop_64:
- decl %ecx
+ decq %rcx
movq %rax,(%rdi)
movq %rax,8(%rdi)
movq %rax,16(%rdi)
@@ -71,7 +95,7 @@ ENTRY(__memset)
to predict jump tables. */
.p2align 4
.Lhandle_tail:
- movl %r11d,%ecx
+ movl %edx,%ecx
andl $63&(~7),%ecx
jz .Lhandle_7
shrl $3,%ecx
@@ -83,12 +107,11 @@ ENTRY(__memset)
jnz .Lloop_8
.Lhandle_7:
- movl %r11d,%ecx
- andl $7,%ecx
+ andl $7,%edx
jz .Lende
.p2align 4
.Lloop_1:
- decl %ecx
+ decl %edx
movb %al,(%rdi)
leaq 1(%rdi),%rdi
jnz .Lloop_1
@@ -99,29 +122,33 @@ ENTRY(__memset)
CFI_RESTORE_STATE
.Lbad_alignment:
- cmpq $7,%r11
+ cmpq $7,%rdx
jbe .Lhandle_7
movq %rax,(%rdi) /* unaligned store */
movq $8,%r8
subq %r9,%r8
addq %r8,%rdi
- subq %r8,%r11
+ subq %r8,%rdx
jmp .Lafter_bad_alignment
.Lfinal:
CFI_ENDPROC
ENDPROC(memset)
ENDPROC(__memset)
- /* Some CPUs run faster using the string instructions.
- It is also a lot simpler. Use this when possible */
-
-#include <asm/cpufeature.h>
-
+ /* Some CPUs support enhanced REP MOVSB/STOSB feature.
+ * It is recommended to use this when possible.
+ *
+ * If enhanced REP MOVSB/STOSB feature is not available, use fast string
+ * instructions.
+ *
+ * Otherwise, use original memset function.
+ *
+ * In .altinstructions section, ERMS feature is placed after REG_GOOD
+ * feature to implement the right patch order.
+ */
.section .altinstructions,"a"
- .align 8
- .quad memset
- .quad .Lmemset_c
- .word X86_FEATURE_REP_GOOD
- .byte .Lfinal - memset
- .byte .Lmemset_e - .Lmemset_c
+ altinstruction_entry memset,.Lmemset_c,X86_FEATURE_REP_GOOD,\
+ .Lfinal-memset,.Lmemset_e-.Lmemset_c
+ altinstruction_entry memset,.Lmemset_c_e,X86_FEATURE_ERMS, \
+ .Lfinal-memset,.Lmemset_e_e-.Lmemset_c_e
.previous