x86-64, mem: Convert memmove() to assembly file and fix return value bug

memmove_64.c only implements memmove() function which is completely written in
inline assembly code. Therefore it doesn't make sense to keep the assembly code
in .c file.

Currently memmove() doesn't store return value to rax. This may cause issue if
caller uses the return value. The patch fixes this issue.

Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
LKML-Reference: <1295314755-6625-1-git-send-email-fenghua.yu@intel.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>

2 files changed, 197 insertions, 192 deletions

diff --git a/arch/x86/lib/memmove_64.S b/arch/x86/lib/memmove_64.S
new file mode 100644
index 000000000000..0ecb8433e5a8
--- /dev/null
+++ b/arch/x86/lib/memmove_64.S
@@ -0,0 +1,197 @@
+/*
+ * Normally compiler builtins are used, but sometimes the compiler calls out
+ * of line code. Based on asm-i386/string.h.
+ *
+ * This assembly file is re-written from memmove_64.c file.
+ *	- Copyright 2011 Fenghua Yu <fenghua.yu@intel.com>
+ */
+#define _STRING_C
+#include <linux/linkage.h>
+#include <asm/dwarf2.h>
+
+#undef memmove
+
+/*
+ * Implement memmove(). This can handle overlap between src and dst.
+ *
+ * Input:
+ * rdi: dest
+ * rsi: src
+ * rdx: count
+ *
+ * Output:
+ * rax: dest
+ */
+ENTRY(memmove)
+	CFI_STARTPROC
+	/* Handle more 32bytes in loop */
+	mov %rdi, %rax
+	cmp $0x20, %rdx
+	jb	1f
+
+	/* Decide forward/backward copy mode */
+	cmp %rdi, %rsi
+	jb	2f
+
+	/*
+	 * movsq instruction have many startup latency
+	 * so we handle small size by general register.
+	 */
+	cmp  $680, %rdx
+	jb	3f
+	/*
+	 * movsq instruction is only good for aligned case.
+	 */
+
+	cmpb %dil, %sil
+	je 4f
+3:
+	sub $0x20, %rdx
+	/*
+	 * We gobble 32byts forward in each loop.
+	 */
+5:
+	sub $0x20, %rdx
+	movq 0*8(%rsi), %r11
+	movq 1*8(%rsi), %r10
+	movq 2*8(%rsi), %r9
+	movq 3*8(%rsi), %r8
+	leaq 4*8(%rsi), %rsi
+
+	movq %r11, 0*8(%rdi)
+	movq %r10, 1*8(%rdi)
+	movq %r9, 2*8(%rdi)
+	movq %r8, 3*8(%rdi)
+	leaq 4*8(%rdi), %rdi
+	jae 5b
+	addq $0x20, %rdx
+	jmp 1f
+	/*
+	 * Handle data forward by movsq.
+	 */
+	.p2align 4
+4:
+	movq %rdx, %rcx
+	movq -8(%rsi, %rdx), %r11
+	lea -8(%rdi, %rdx), %r10
+	shrq $3, %rcx
+	rep movsq
+	movq %r11, (%r10)
+	jmp 13f
+	/*
+	 * Handle data backward by movsq.
+	 */
+	.p2align 4
+7:
+	movq %rdx, %rcx
+	movq (%rsi), %r11
+	movq %rdi, %r10
+	leaq -8(%rsi, %rdx), %rsi
+	leaq -8(%rdi, %rdx), %rdi
+	shrq $3, %rcx
+	std
+	rep movsq
+	cld
+	movq %r11, (%r10)
+	jmp 13f
+
+	/*
+	 * Start to prepare for backward copy.
+	 */
+	.p2align 4
+2:
+	cmp $680, %rdx
+	jb 6f
+	cmp %dil, %sil
+	je 7b
+6:
+	/*
+	 * Calculate copy position to tail.
+	 */
+	addq %rdx, %rsi
+	addq %rdx, %rdi
+	subq $0x20, %rdx
+	/*
+	 * We gobble 32byts backward in each loop.
+	 */
+8:
+	subq $0x20, %rdx
+	movq -1*8(%rsi), %r11
+	movq -2*8(%rsi), %r10
+	movq -3*8(%rsi), %r9
+	movq -4*8(%rsi), %r8
+	leaq -4*8(%rsi), %rsi
+
+	movq %r11, -1*8(%rdi)
+	movq %r10, -2*8(%rdi)
+	movq %r9, -3*8(%rdi)
+	movq %r8, -4*8(%rdi)
+	leaq -4*8(%rdi), %rdi
+	jae 8b
+	/*
+	 * Calculate copy position to head.
+	 */
+	addq $0x20, %rdx
+	subq %rdx, %rsi
+	subq %rdx, %rdi
+1:
+	cmpq $16, %rdx
+	jb 9f
+	/*
+	 * Move data from 16 bytes to 31 bytes.
+	 */
+	movq 0*8(%rsi), %r11
+	movq 1*8(%rsi), %r10
+	movq -2*8(%rsi, %rdx), %r9
+	movq -1*8(%rsi, %rdx), %r8
+	movq %r11, 0*8(%rdi)
+	movq %r10, 1*8(%rdi)
+	movq %r9, -2*8(%rdi, %rdx)
+	movq %r8, -1*8(%rdi, %rdx)
+	jmp 13f
+	.p2align 4
+9:
+	cmpq $8, %rdx
+	jb 10f
+	/*
+	 * Move data from 8 bytes to 15 bytes.
+	 */
+	movq 0*8(%rsi), %r11
+	movq -1*8(%rsi, %rdx), %r10
+	movq %r11, 0*8(%rdi)
+	movq %r10, -1*8(%rdi, %rdx)
+	jmp 13f
+10:
+	cmpq $4, %rdx
+	jb 11f
+	/*
+	 * Move data from 4 bytes to 7 bytes.
+	 */
+	movl (%rsi), %r11d
+	movl -4(%rsi, %rdx), %r10d
+	movl %r11d, (%rdi)
+	movl %r10d, -4(%rdi, %rdx)
+	jmp 13f
+11:
+	cmp $2, %rdx
+	jb 12f
+	/*
+	 * Move data from 2 bytes to 3 bytes.
+	 */
+	movw (%rsi), %r11w
+	movw -2(%rsi, %rdx), %r10w
+	movw %r11w, (%rdi)
+	movw %r10w, -2(%rdi, %rdx)
+	jmp 13f
+12:
+	cmp $1, %rdx
+	jb 13f
+	/*
+	 * Move data for 1 byte.
+	 */
+	movb (%rsi), %r11b
+	movb %r11b, (%rdi)
+13:
+	retq
+	CFI_ENDPROC
+ENDPROC(memmove)
diff --git a/arch/x86/lib/memmove_64.c b/arch/x86/lib/memmove_64.c
deleted file mode 100644
index 6d0f0ec41b34..000000000000
--- a/arch/x86/lib/memmove_64.c
+++ /dev/null
@@ -1,192 +0,0 @@
-/* Normally compiler builtins are used, but sometimes the compiler calls out
-   of line code. Based on asm-i386/string.h.
- */
-#define _STRING_C
-#include <linux/string.h>
-#include <linux/module.h>
-
-#undef memmove
-void *memmove(void *dest, const void *src, size_t count)
-{
-	unsigned long d0,d1,d2,d3,d4,d5,d6,d7;
-	char *ret;
-
-	__asm__ __volatile__(
-		/* Handle more 32bytes in loop */
-		"mov %2, %3\n\t"
-		"cmp $0x20, %0\n\t"
-		"jb	1f\n\t"
-
-		/* Decide forward/backward copy mode */
-		"cmp %2, %1\n\t"
-		"jb	2f\n\t"
-
-		/*
-		 * movsq instruction have many startup latency
-		 * so we handle small size by general register.
-		 */
-		"cmp  $680, %0\n\t"
-		"jb 3f\n\t"
-		/*
-		 * movsq instruction is only good for aligned case.
-		 */
-		"cmpb %%dil, %%sil\n\t"
-		"je 4f\n\t"
-		"3:\n\t"
-		"sub $0x20, %0\n\t"
-		/*
-		 * We gobble 32byts forward in each loop.
-		 */
-		"5:\n\t"
-		"sub $0x20, %0\n\t"
-		"movq 0*8(%1), %4\n\t"
-		"movq 1*8(%1), %5\n\t"
-		"movq 2*8(%1), %6\n\t"
-		"movq 3*8(%1), %7\n\t"
-		"leaq 4*8(%1), %1\n\t"
-
-		"movq %4, 0*8(%2)\n\t"
-		"movq %5, 1*8(%2)\n\t"
-		"movq %6, 2*8(%2)\n\t"
-		"movq %7, 3*8(%2)\n\t"
-		"leaq 4*8(%2), %2\n\t"
-		"jae 5b\n\t"
-		"addq $0x20, %0\n\t"
-		"jmp 1f\n\t"
-		/*
-		 * Handle data forward by movsq.
-		 */
-		".p2align 4\n\t"
-		"4:\n\t"
-		"movq %0, %8\n\t"
-		"movq -8(%1, %0), %4\n\t"
-		"lea -8(%2, %0), %5\n\t"
-		"shrq $3, %8\n\t"
-		"rep movsq\n\t"
-		"movq %4, (%5)\n\t"
-		"jmp 13f\n\t"
-		/*
-		 * Handle data backward by movsq.
-		 */
-		".p2align 4\n\t"
-		"7:\n\t"
-		"movq %0, %8\n\t"
-		"movq (%1), %4\n\t"
-		"movq %2, %5\n\t"
-		"leaq -8(%1, %0), %1\n\t"
-		"leaq -8(%2, %0), %2\n\t"
-		"shrq $3, %8\n\t"
-		"std\n\t"
-		"rep movsq\n\t"
-		"cld\n\t"
-		"movq %4, (%5)\n\t"
-		"jmp 13f\n\t"
-
-		/*
-		 * Start to prepare for backward copy.
-		 */
-		".p2align 4\n\t"
-		"2:\n\t"
-		"cmp $680, %0\n\t"
-		"jb 6f \n\t"
-		"cmp %%dil, %%sil\n\t"
-		"je 7b \n\t"
-		"6:\n\t"
-		/*
-		 * Calculate copy position to tail.
-		 */
-		"addq %0, %1\n\t"
-		"addq %0, %2\n\t"
-		"subq $0x20, %0\n\t"
-		/*
-		 * We gobble 32byts backward in each loop.
-		 */
-		"8:\n\t"
-		"subq $0x20, %0\n\t"
-		"movq -1*8(%1), %4\n\t"
-		"movq -2*8(%1), %5\n\t"
-		"movq -3*8(%1), %6\n\t"
-		"movq -4*8(%1), %7\n\t"
-		"leaq -4*8(%1), %1\n\t"
-
-		"movq %4, -1*8(%2)\n\t"
-		"movq %5, -2*8(%2)\n\t"
-		"movq %6, -3*8(%2)\n\t"
-		"movq %7, -4*8(%2)\n\t"
-		"leaq -4*8(%2), %2\n\t"
-		"jae 8b\n\t"
-		/*
-		 * Calculate copy position to head.
-		 */
-		"addq $0x20, %0\n\t"
-		"subq %0, %1\n\t"
-		"subq %0, %2\n\t"
-		"1:\n\t"
-		"cmpq $16, %0\n\t"
-		"jb 9f\n\t"
-		/*
-		 * Move data from 16 bytes to 31 bytes.
-		 */
-		"movq 0*8(%1), %4\n\t"
-		"movq 1*8(%1), %5\n\t"
-		"movq -2*8(%1, %0), %6\n\t"
-		"movq -1*8(%1, %0), %7\n\t"
-		"movq %4, 0*8(%2)\n\t"
-		"movq %5, 1*8(%2)\n\t"
-		"movq %6, -2*8(%2, %0)\n\t"
-		"movq %7, -1*8(%2, %0)\n\t"
-		"jmp 13f\n\t"
-		".p2align 4\n\t"
-		"9:\n\t"
-		"cmpq $8, %0\n\t"
-		"jb 10f\n\t"
-		/*
-		 * Move data from 8 bytes to 15 bytes.
-		 */
-		"movq 0*8(%1), %4\n\t"
-		"movq -1*8(%1, %0), %5\n\t"
-		"movq %4, 0*8(%2)\n\t"
-		"movq %5, -1*8(%2, %0)\n\t"
-		"jmp 13f\n\t"
-		"10:\n\t"
-		"cmpq $4, %0\n\t"
-		"jb 11f\n\t"
-		/*
-		 * Move data from 4 bytes to 7 bytes.
-		 */
-		"movl (%1), %4d\n\t"
-		"movl -4(%1, %0), %5d\n\t"
-		"movl %4d, (%2)\n\t"
-		"movl %5d, -4(%2, %0)\n\t"
-		"jmp 13f\n\t"
-		"11:\n\t"
-		"cmp $2, %0\n\t"
-		"jb 12f\n\t"
-		/*
-		 * Move data from 2 bytes to 3 bytes.
-		 */
-		"movw (%1), %4w\n\t"
-		"movw -2(%1, %0), %5w\n\t"
-		"movw %4w, (%2)\n\t"
-		"movw %5w, -2(%2, %0)\n\t"
-		"jmp 13f\n\t"
-		"12:\n\t"
-		"cmp $1, %0\n\t"
-		"jb 13f\n\t"
-		/*
-		 * Move data for 1 byte.
-		 */
-		"movb (%1), %4b\n\t"
-		"movb %4b, (%2)\n\t"
-		"13:\n\t"
-		: "=&d" (d0), "=&S" (d1), "=&D" (d2), "=&a" (ret) ,
-		  "=r"(d3), "=r"(d4), "=r"(d5), "=r"(d6), "=&c" (d7)
-		:"0" (count),
-		 "1" (src),
-		 "2" (dest)
-		:"memory");
-
-		return ret;
-
-}
-EXPORT_SYMBOL(memmove);