x86, mem: memmove_64.S: Optimize memmove by enhanced REP MOVSB/STOSB

Support memmove() by enhanced rep movsb. On processors supporting enhanced
REP MOVSB/STOSB, the alternative memmove() function using enhanced rep movsb
overrides the original function.

The patch doesn't change the backward memmove case to use enhanced rep
movsb.

Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
Link: http://lkml.kernel.org/r/1305671358-14478-9-git-send-email-fenghua.yu@intel.com
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>

arch/x86/lib/memmove_64.S

@@ -8,6 +8,7 @@
 #define _STRING_C
 #include <linux/linkage.h>
 #include <asm/dwarf2.h>
+#include <asm/cpufeature.h>
 
 #undef memmove
 
@@ -24,6 +25,7 @@
  */
 ENTRY(memmove)
 	CFI_STARTPROC
+
 	/* Handle more 32bytes in loop */
 	mov %rdi, %rax
 	cmp $0x20, %rdx
@@ -31,8 +33,13 @@ ENTRY(memmove)
 
 	/* Decide forward/backward copy mode */
 	cmp %rdi, %rsi
-	jb	2f
+	jge .Lmemmove_begin_forward
+	mov %rsi, %r8
+	add %rdx, %r8
+	cmp %rdi, %r8
+	jg 2f
 
+.Lmemmove_begin_forward:
 	/*
 	 * movsq instruction have many startup latency
 	 * so we handle small size by general register.
@@ -78,6 +85,8 @@ ENTRY(memmove)
 	rep movsq
 	movq %r11, (%r10)
 	jmp 13f
+.Lmemmove_end_forward:
+
 	/*
 	 * Handle data backward by movsq.
 	 */
@@ -194,4 +203,22 @@ ENTRY(memmove)
 13:
 	retq
 	CFI_ENDPROC
+
+	.section .altinstr_replacement,"ax"
+.Lmemmove_begin_forward_efs:
+	/* Forward moving data. */
+	movq %rdx, %rcx
+	rep movsb
+	retq
+.Lmemmove_end_forward_efs:
+	.previous
+
+	.section .altinstructions,"a"
+	.align 8
+	.quad .Lmemmove_begin_forward
+	.quad .Lmemmove_begin_forward_efs
+	.word X86_FEATURE_ERMS
+	.byte .Lmemmove_end_forward-.Lmemmove_begin_forward
+	.byte .Lmemmove_end_forward_efs-.Lmemmove_begin_forward_efs
+	.previous
 ENDPROC(memmove)